security
97 pieces
Account Recovery Is Your Real Password: How to Audit It
Attackers choose the weakest way back into an account, and that is rarely the login screen.
Most users treat account recovery security as an afterthought, leaving doors open for attackers who bypass strong login credentials entirely. A simple audit of recovery paths on your primary accounts closes these gaps before they are exploited.
2026-09-14 · technical-essay · 8 min read
AI Coding Assistants and Leaked API Keys: Where Secrets Escape
Secrets now leak through context windows and logs, not only through commits.
The ai coding assistant api key leak is no longer a risk limited to accidental commits. Agents read the entire workspace, copying .env files into prompts and logs before any version control system can intervene. You must isolate secrets from the agent's view and use short-lived tokens to close these new paths.
2026-09-14 · technical-essay · 9 min read
AI in the Security Operations Center: What It Actually Speeds Up
Summaries and triage help analysts; autonomous response amplifies every false positive.
AI in soc delivers value through summarisation and context gathering, not autonomous action. Letting models handle containment turns detection errors into outages. Human approval must remain on any action that disconnects or disables systems.
2026-09-14 · technical-essay · 9 min read
AI Red Teaming for Small Teams: A One-Week Plan
Test what your AI features can do when fed hostile content, not how rude a jailbreak can make them.
Small teams often misdirect their efforts in ai red teaming by chasing generic jailbreaks that test the model vendor rather than their own product. This plan shifts focus to planting benign canary instructions in every content source to measure actual data leakage and tool access. The result is a practical security assessment that reveals real risks to your specific architecture.
2026-09-14 · technical-essay · 9 min read
An AI Agent Kill Switch: Design the Stop Button First
Most deployments cannot halt one agent mid-task without killing everything else.
Deployments rarely prioritise how quickly a human can stop an agent. A usable ai agent kill switch must be out-of-band and revoke credentials rather than ask the agent to stop. This design choice ensures governance before the first tool is connected.
2026-09-14 · field-note · 8 min read
An Incident Response Plan That Fits on One Page
Plans fail on who decides and who to call, not on the technical steps.
Most incident response plan template documents gather dust because they prioritise procedure over authority. A one-page sheet that names the decision-maker and external contacts prevents delays when systems are down. This approach minimises confusion and ensures evidence is preserved before cleanup begins.
2026-09-14 · technical-essay · 7 min read
Are AI Browsers Safe? The Logged-In Session Problem
An agent that browses as you inherits every account you are signed in to.
The question of whether are ai browsers safe depends less on the model’s intelligence and more on the permissions it inherits. When an agent operates within your authenticated session, it gains access to every email, bank account, and work tool you use daily. The practical safeguard is to isolate agent tasks in a separate profile with no standing logins to sensitive services.
2026-09-14 · technical-essay · 9 min read
Broken Object Level Authorization: The API Flaw Behind Big Leaks
Change one ID in a request and read someone else's records: the most common API failure.
Broken object level authorization allows attackers to access data belonging to other users by simply altering identifiers in API requests. This vulnerability persists because many frameworks separate authentication from authorisation, leaving ownership checks to individual endpoints. Fixing it requires structural changes to how systems verify user permissions.
2026-09-14 · technical-essay · 9 min read
Can AI Write Malware? What Has Actually Changed for Attackers
Language models cut the cost of variation and lures far more than they invent new attacks.
The question can ai write malware misses the point. Models do not invent new exploits. They automate the tedious work of variation and persuasion. This shifts the burden from detection to behavioural analysis.
2026-09-14 · technical-essay · 9 min read
Cloud Misconfiguration: Why Exposed Storage Keeps Happening
Drift and convenience defeat one-time audits; guardrails have to sit above individual accounts.
Cloud misconfiguration persists not from negligence but from the friction between rapid deployment and static controls. When infrastructure changes faster than policy, convenience wins. Sustainable security requires organisational guardrails that make incorrect states impossible, rather than relying on audits that miss the next change.
2026-09-14 · technical-essay · 8 min read
Crypto Wallet Drainers: The Signature You Did Not Read
Drainers rarely steal seed phrases; they get you to sign a permission that outlives the session.
A crypto wallet drainer does not need your seed phrase to empty your account. It relies on you signing a transaction that grants ongoing transfer rights. Understanding how these permissions work is the only reliable defence.
2026-09-14 · technical-essay · 8 min read
Cyber Incident Reporting Deadlines: Several Clocks, One Incident
A single breach can trigger 24-hour, 72-hour and materiality-based deadlines to different regulators.
Organisations often anchor on the GDPR 72-hour window and miss earlier triggers. Mapping the specific events that start each clock matters more than memorising durations. Clear cyber incident reporting requirements prevent regulatory penalties and reputational damage.
2026-09-14 · technical-essay · 9 min read
Deepfake Job Candidates: Remote Hiring Is Now an Identity Problem
The interview has become an authentication step, and most hiring teams never designed it as one.
The rise of deepfake job candidates reveals a fundamental flaw in remote hiring. Interviews now function as authentication steps that most organisations have not secured. We must treat identity assurance as a security control, not a recruiting formality.
2026-09-14 · technical-essay · 9 min read
Digital Legacy: Planning Your Accounts and Photos After Death
Setting up legacy contacts and password inheritance ensures your family can access or close accounts without legal gridlock.
Digital legacy planning prevents your family from being locked out of essential accounts and memories. Proactive configuration of legacy contacts and secure password sharing is the only reliable method to manage posthumous access. Without it, legal processes often fail to resolve technical barriers.
2026-09-14 · technical-essay · 9 min read
Fake CAPTCHA Scams: Never Paste What a Website Tells You To
ClickFix pages turn the visitor into the installer, bypassing every filter in between.
A fake captcha scam tricks users into pasting malicious commands into their own systems. This clickfix attack exploits trust in verification steps. The solution is simple: never open command prompts on instruction from a browser.
2026-09-14 · technical-essay · 10 min read
GPS Spoofing Explained: When Location Data Lies
Jamming announces itself; spoofing keeps the map calm and confident while it moves you somewhere else.
Satellite positioning was built to be received by anyone, not authenticated. This means a receiver has no native way to tell a real signal from a convincing fake. Understanding what is gps spoofing reveals why location data is fragile and why cross-checking is essential for security.
2026-09-14 · technical-essay · 7 min read
Help Desk Social Engineering: The Password Reset Phone Call
A convincing caller asking IT for a reset has become the easiest way past strong authentication.
Strong authentication fails when the help desk resets it for anyone who sounds right. This analysis examines how help desk social engineering bypasses technical controls through voice manipulation and procedural gaps, outlining verification methods that do not rely on public facts.
2026-09-14 · technical-essay · 8 min read
Home Router Security: The Settings That Don't
Hiding the network name changes nothing; the admin password, updates and remote management change almost everything.
Most people waste time on rituals that feel secure while ignoring the controls that actually matter. Effective home router security settings focus on access control, patching, and isolation rather than cosmetic changes. This guide separates signal from noise to protect your network.
2026-09-14 · technical-essay · 10 min read
How to Know If Your Phone Is Hacked: Signs That Actually Matter
Battery drain and heat are mostly noise; account-level evidence is where compromise shows.
Most people chase phantom symptoms like battery drain when the real evidence sits in their accounts. You can determine how to know if your phone is hacked by checking for unfamiliar sessions and changed recovery details. This approach is far more reliable than scanning for spyware.
2026-09-14 · technical-essay · 8 min read
Infostealer Malware: Why a Stolen Cookie Beats a Stolen Password
After an infostealer, changing passwords is not enough; sessions must be killed everywhere.
Infostealer malware captures active session tokens, allowing attackers to bypass passwords and two-factor authentication entirely. Changing credentials without revoking these sessions leaves accounts vulnerable. Effective recovery requires cleaning the device and signing out everywhere before updating any secrets.
2026-09-14 · technical-essay · 9 min read
Initial Access Brokers: The Market Behind a Ransomware Attack
The intrusion you suffer was often bought, and your leaked logins set the price.
Many ransomware incidents begin with a transaction rather than a hack. Initial access brokers sell working access harvested from infostealer logs or exposed remote services. Organisations can measure part of their risk from outside by watching for their own credentials and remote-access exposure.
2026-09-14 · technical-essay · 8 min read
Insider Threat in Remote Teams: Signals Without Surveillance
Watching access and data movement works better than watching employees.
Most insider threat programmes fail because they watch people instead of data. Screen recording erodes trust and misses the actual exfiltration. We must shift focus to access patterns and data movement to catch real risks without invasive surveillance.
2026-09-14 · technical-essay · 9 min read
Internet-Exposed PLCs: Why Small Utilities Keep Getting Hit
Attacks on water systems exploit defaults and exposure, not sophistication.
Intrusions into small utilities generally rely on controllers reachable from the internet with default or shared credentials. This makes them a funding and ownership problem more than a technical mystery. The protective moves are known and cheap relative to the harm.
2026-09-14 · technical-essay · 8 min read
Juice Jacking: Is Public USB Charging Actually Dangerous?
The famous warning is mostly theory; the smaller real risk is tapping Trust on a prompt.
The concept of juice jacking is widely warned about but rarely documented in the wild. Treating it as a primary travel threat displaces attention from more common risks. The sensible approach is simple: use your own charger or a power-only adapter, and never accept a data prompt on an unknown device.
2026-09-14 · technical-essay · 8 min read
Keyless Car Theft: Relay Attacks and What Actually Stops Them
A signal-blocking pouch works only on the nights you remember it; layers the relay cannot reach work every night.
A keyless car theft relay attack does not break cryptography; it lies about distance. Habit-based defences like pouches fail when forgotten. Robust security requires removing the signal by default or adding physical barriers the relay cannot carry.
2026-09-14 · technical-essay · 9 min read
MCP Server Security: A Config File Is Now Executable Code
Cloning a repository can launch tools with your privileges before you have read a line.
Project-level MCP configuration turns a JSON file into a process launcher that runs with your operating-system privileges. The trust decision has moved from running code to opening a folder, requiring teams to treat these configs with the same rigour as build scripts. This article examines the mechanisms behind mcp server security and how to mitigate the risks.
2026-09-14 · technical-essay · 9 min read
Non-Human Identities: The Credentials Nobody Owns
Service accounts, API keys and tokens outnumber people, and most have no owner or expiry.
Most organisations treat machine credentials as disposable, yet they persist long after their purpose fades. Non-human identities lack the natural offboarding cycle of human employees, creating a silent accumulation of access. The solution is not more tools, but strict ownership and expiry rules for every credential.
2026-09-14 · technical-essay · 8 min read
Phishing Kits That Bypass MFA: A Subscription Business
Adversary-in-the-middle phishing is sold with dashboards, updates and support.
Adversaries now rent proxy kits that capture full sessions after MFA login, collapsing the skill barrier for bypassing authentication. This shift from one-off exploits to a subscription service demands a fundamental change in how we design identity systems.
2026-09-14 · technical-essay · 8 min read
Phone Stolen? The First 30 Minutes, in Order
Thieves who watched your passcode are after your accounts, not the handset.
When your phone is stolen, the hardware is secondary to the digital identity it holds. Understanding phone stolen what to do requires prioritising account recovery over device tracking. Thieves with your passcode can bypass local locks to access cloud backups and reset passwords.
2026-09-14 · technical-essay · 8 min read
QR Code Scams (Quishing): Why a Sticker Beats Your Spam Filter
QR codes move a link out of screened channels and onto an unmanaged phone, often in the physical world.
Quishing exploits the blind spot between digital security and physical reality. By moving phishing links from screened email inboxes to unmanaged mobile cameras, attackers bypass traditional defences. The solution is not more software, but a shift to contextual verification and manual address entry.
2026-09-14 · technical-essay · 9 min read
Ransomware as a Service: How the Criminal Franchise Works
Brands come and go; affiliates, brokers and negotiators stay in business.
Ransomware as a service operates like a criminal franchise, separating tool development from execution and laundering. This division of labour ensures that takedowns disrupt brands but rarely dismantle the underlying economy. Defenders must analyse affiliate behaviour rather than chasing specific group names to understand the threat.
2026-09-14 · technical-essay · 8 min read
Running a Local LLM for Privacy: What You Gain and What You Don't
Local models remove the provider from the picture, not the risks of what the model is allowed to do.
Running a local llm privacy is often misunderstood as a total security solution. While it stops data leaving your device, it does not protect against prompt injection or supply chain risks. Local deployment is a privacy choice, not a comprehensive security strategy.
2026-09-14 · field-note · 8 min read
Setting Up a Child's First Phone: Contact Controls Before Content
The biggest risk on a first phone is who can reach the child, not what they might see.
Most parents focus on content filters for their child's first phone safety settings, yet the fastest escalation of harm comes through contact. Configuring who can message, call, and add the child matters more than any filter list. Prioritising communication limits reduces exposure to grooming and scams before they begin.
2026-09-14 · technical-essay · 8 min read
Should Open-Weight AI Models Be Regulated?
Weights cannot be recalled once published, which forces regulation to look somewhere else.
Open weight ai regulation requires a shift from post-deployment control to pre-release scrutiny. Once parameters are public, recall is impossible and monitoring is ineffective. Policy must therefore target the release process and specific downstream applications rather than the model files themselves.
2026-09-14 · technical-essay · 8 min read
Should You Pay a Ransomware Demand? The Case Against Certainty
Payment buys a promise from a counterparty with no reason to keep it.
The question of should you pay ransomware is not an ethical dilemma but an epistemic failure. Organisations cannot verify the promises made by attackers, making payment a high-risk gamble against a counterparty with no incentive to honour it. Decisions made under duress lack the evidence required for sound judgement.
2026-09-14 · technical-essay · 8 min read
Slopsquatting: When AI Invents a Package, Someone Registers It
Hallucinated dependency names repeat across users and sessions, which makes them predictable targets.
Slopsquatting exploits the reproducibility of AI hallucinations. Attackers register packages that models frequently invent, turning statistical noise into a supply chain attack vector. Defence requires mechanical checks, not just better prompts.
2026-09-14 · technical-essay · 10 min read
Small Business Cybersecurity: The First Five Things, in Order
Before buying any product, a small firm needs five unglamorous controls done properly.
Most small firms buy enterprise tools while ignoring basic hygiene. True small business cybersecurity relies on five foundational controls: strong email authentication, separated admin accounts, tested backups, payment verification, and a contact list. These steps prevent the majority of common attacks without requiring complex software.
2026-09-14 · technical-essay · 8 min read
Vibe Coding Security Risks: The Check That Was Never Written
AI-generated apps rarely fail through exotic bugs; they fail through omissions nobody asked for.
Vibe coding security risks stem not from complex exploits but from missing safeguards. AI models generate happy paths by default, leaving out authentication and input validation. This essay argues that reviewing such code requires auditing for absences rather than hunting for errors.
2026-09-14 · technical-essay · 8 min read
What Is Prompt Injection? Why It Is Not Like SQL Injection
The comparison suggests a patch exists; the architecture says otherwise.
Understanding what is prompt injection requires abandoning the SQL injection analogy. Language models process instructions and data in a single stream, making simple separation impossible. Durable security relies on limiting consequences rather than detecting malicious intent.
2026-09-14 · technical-essay · 9 min read
What LLM Guardrails Can and Cannot Stop
A probabilistic filter facing an attacker with unlimited retries is a rate reducer, not a wall.
LLM guardrails reduce the rate of bad outcomes but fail against determined attackers who iterate through attempts. They are valuable tools only when the system remains safe even if the filter misses. This analysis explains their mechanical limits and how to design for failure.
2026-09-14 · technical-essay · 9 min read
When AI Agents Talk to Each Other, Who Is in Charge?
Delegation chains can launder a stranger's instruction into something that looks authorised.
In multi-agent ai security, the core risk is not just failure, but the silent laundering of authority. When agents delegate tasks without carrying the original intent, a stranger's command can masquerade as a trusted directive, bypassing the safeguards designed to protect privileged actions.
2026-09-14 · technical-essay · 9 min read
When Ransomware Hits a Hospital, It Is a Patient Safety Event
Treating a hospital cyberattack as an IT outage underprepares the clinicians who carry it.
Hospital ransomware patient safety risks are often obscured by technical recovery timelines. Clinicians face immediate care degradation when digital systems fail. This essay argues that cyber incidents must be managed as clinical emergencies, not just IT outages.
2026-09-14 · technical-essay · 8 min read
Will Quantum Computers Break Encryption? What Is Actually at Risk
Public-key exchange and signatures are exposed; your AES files and hashed passwords mostly are not.
Popular coverage implies quantum computers will unlock everything, but the realistic exposure is concentrated in public-key algorithms. Symmetric encryption and hashing need larger parameters rather than replacement. The ordinary person's most useful action is keeping software updated rather than worrying.
2026-09-14 · technical-essay · 8 min read
Zero Trust Explained Without the Vendor Pitch
Zero trust is a set of per-request decisions, and one stolen laptop tests whether you have it.
Many ask what is zero trust, expecting a product to buy. It is actually a discipline of per-request verification. This essay explains the architecture without the vendor pitch, focusing on practical defence mechanisms.
2026-09-14 · technical-essay · 8 min read
Evasion Attacks via Adversarial Perturbations
The failure is confident, which is what removes the signal that would have caught it
A classifier can be made to fail on a change too small for a person to notice, and it fails without hesitation. Why that confidence is the real damage, where the risk is concentrated, and why the durable answer is architectural rather than a better model.
2026-09-13 · technical-essay · 4 min read
Improper Output Handling
The moment generated code reaches a shell, it stops mattering who wrote it
Code a model produced is code from an untrusted author, and the review step that would exist for a person's work is missing because the output arrived in milliseconds looking finished. Why plausibility is the trap, and why a sandbox bounds damage without saying anything about correctness.
2026-09-13 · technical-essay · 4 min read
Insecure Plugin Integrations
The assistant became the place where two estates meet, and nobody negotiated the terms
Connecting an assistant to an external tool extends your trust boundary to an operator you have no agreement with, at a seam authenticated more weakly than anything else you run. What the integration can actually see, and the line where the problem stops being technical.
2026-09-13 · technical-essay · 3 min read
Supply Chain Compromises in Open-Source Model Repositories
Copying an identifier into a config is a procurement decision that nobody recorded
Pulling a model from a public hub commits you to an uploader, a namespace and everything pushed to it since, none of which was assessed. Why a name is not a version, why some weight formats execute on load, and what provenance can and cannot tell you.
2026-09-13 · technical-essay · 4 min read
'Military-Grade Encryption' Means Nothing
A guide to the security vocabulary that sounds strongest and tells you least
Some security claims describe a property that can be checked. Others describe a feeling. Sorting the phrases in common marketing use into those that carry information and those that carry only reassurance, with the question that exposes each one.
2026-09-12 · technical-essay · 3 min read
Age Verification and the Identity Trap
Proving you are over eighteen usually means proving exactly who you are, and the two are not the same requirement
A system that checks age by collecting identity documents has answered a yes-or-no question by building a register. The gap between what is being asked and what is being collected, why implementations default to the wider one, and what a narrow answer would look like.
2026-09-12 · technical-essay · 3 min read
AI Detectors Do Not Work, and the Cost Is Not Evenly Shared
Why classifying text as machine-written is structurally hard, and who pays when an institution pretends otherwise
Tools claiming to identify machine-written text are deployed in schools, universities and hiring, on the assumption that they are approximately right. The reasons they cannot be reliable are structural rather than temporary, and the errors fall hardest on people least able to contest them.
2026-09-12 · technical-essay · 4 min read
Anonymised Is a Verb, Not a State
Removing names is the easy part, and it is almost never the part that identifies you
Organisations describe data as anonymised after removing direct identifiers, then release or trade it as though identification were now impossible. What actually identifies a person in a dataset, why re-identification succeeds so reliably, and what the word would have to mean to be worth anything.
2026-09-12 · technical-essay · 3 min read
API Parameter Exploitation in Hybrid AI Models
The trusted line is the one nobody drew
Conventional injection flaws re-enter systems at the point where a model's output rejoins an API pipeline, because that output is treated as internal when it never was. The reader leaves with a working boundary rule: validate at every crossing, not just at the front door.
2026-09-12 · technical-essay · 4 min read
Building for the Case Where You Are Compromised
Most systems are designed to prevent a breach. The interesting question is what they are worth after one.
Prevention eventually fails, and the design decisions that matter are the ones determining how much an attacker gets when it does. A way of thinking about architecture that starts from the assumption of compromise rather than treating it as the failure case.
2026-09-12 · founder-essay · 4 min read
Data Poisoning in Fine-Tuning Pipelines
A few crafted examples bend behaviour on a narrow slice while every aggregate stays flat
Evaluation scores cannot see a targeted change, because a targeted change is designed not to move a mean. Why steering a model is a more useful attack than degrading one, and why the defence is provenance rather than a better benchmark.
2026-09-12 · technical-essay · 3 min read
Deleting Your Account Rarely Deletes You
What the delete button reaches, what it cannot reach, and why the difference is structural
Account deletion removes your access and usually your profile. It does not reach backups, derived data, systems that already copied the record, or the conclusions drawn from it. What actually happens after the confirmation dialogue, and the one case where deletion genuinely works.
2026-09-12 · technical-essay · 3 min read
Encrypted at Rest Is the Weakest Claim on the Page
It defends against one specific event, it is nearly universal, and it is presented as though it were the headline
Almost every service states that data is encrypted at rest. The claim is usually true and protects against a narrow scenario: somebody obtaining the physical medium. Against the threats people are actually worried about, it does approximately nothing, and understanding why clarifies what to look for instead.
2026-09-12 · technical-essay · 3 min read
How to Read a Breach Notification
The letter is a carefully constructed document, and the interesting information is in what it declines to say
Breach notifications follow a recognisable template built under legal advice and time pressure. Reading one properly means attending to the tense of the verbs, the scope of the nouns, and the questions the letter answers instead of the ones you asked. A guide to the standard phrases and what each one leaves open.
2026-09-12 · technical-essay · 4 min read
How to Read an Interface
A method for working out what a product optimises, using nothing but a stopwatch and a count of taps
You can infer what an organisation measures from the layout it ships, without access to anybody's intentions or documents. Five signals to look at, why asymmetry of effort is the reliable one, and how to state the reading as evidence rather than as an accusation.
2026-09-12 · technical-essay · 4 min read
Indirect Prompt Injection in Enterprise Knowledge Bases
The payload arrives through a sanctioned route, sits inert, and fires on somebody else's question
The attack surface is the document store. A payload enters through a supplier's file or a wiki edit, waits in the index, and is pulled into context by an unrelated query. Why scanning at upload does not catch it, and what treating retrieval as provenanced data actually buys.
2026-09-12 · technical-essay · 4 min read
Metadata Is the Message
Why who you contacted, when, and from where says more than what you said
Content is expensive to analyse and easy to encrypt. Metadata is cheap to analyse, hard to hide, and sufficient for most conclusions anybody wants to draw about a person. Why the distinction is drawn where it is, and what follows from it.
2026-09-12 · technical-essay · 4 min read
Model Extraction and Reverse Engineering
A public endpoint is a slow, lossy, complete description of the thing behind it
Every answer an API returns is a labelled training pair, given away. Why an attacker needs a model that behaves the same rather than the weights themselves, why the features that make an API pleasant are the ones that make it cheap to copy, and why the honest goal is cost rather than prevention.
2026-09-12 · technical-essay · 3 min read
Model Poisoning Against Data Poisoning
Same headline, different door, and most defences are guarding the wrong one
Two attacks share a name but not a mechanism: corrupting what a model learns from, and corrupting the model itself once training has finished. Why the second is the easier attack in practice, and what actually mitigates it.
2026-09-12 · technical-essay · 4 min read
Over-Privileged Autonomous Agents
The credential was broad so the thing would work, and narrowing it now risks breaking a thing that works
Agents end up over-privileged through a dynamic nobody chose: a wide scope is issued during development and never revisited, because revisiting it carries a certain cost against an uncertain one. What the blast radius actually is, why intent is not a control, and what bounds it.
2026-09-12 · technical-essay · 3 min read
Passkeys Without the Marketing
What replaces the password, and the part nobody mentions until you change phones
Passkeys replace a shared secret with a key pair, which removes an entire category of attack at a stroke. They also move the hard problem from remembering to recovery, and the recovery story is where the differences between implementations actually live.
2026-09-12 · technical-essay · 4 min read
Public Wi-Fi: What Actually Changed
The advice everyone repeats was written for an internet that no longer exists
Warnings about coffee shop networks date from a period when most traffic was unencrypted. Nearly all of it is encrypted now, which changes what a hostile network can and cannot do. What the real remaining risks are, and which of the familiar precautions still earn their place.
2026-09-12 · technical-essay · 3 min read
RAG Vector Database Exploitation
The index is filed as infrastructure and behaves as memory, so write access to it is write access to what the system believes
A vector store is usually secured like a database and used like a recollection. Why permissions applied at the document store do not travel into the index, why the embedding often outlives the file it came from, and what signing and re-embedding actually fix.
2026-09-12 · technical-essay · 3 min read
System Prompt Leakage
Text kept as configuration, which behaves as documentation the moment it escapes
A leaked system prompt is not an embarrassment, it is a map — tool names, argument shapes, role names and refusal conditions, which together mark the edges worth probing. Why it cannot be kept secret, and how to make the leak boring instead.
2026-09-12 · technical-essay · 3 min read
The Backup You Have Not Tested Does Not Exist
Backup is a verb about restoring, and almost everybody has only done the first half
Most backup arrangements have never been used. The failure modes that emerge only on the day you need them, why the encryption question is sharper here than anywhere else, and a test that takes an hour and settles it.
2026-09-12 · technical-essay · 4 min read
The Cost of Being Findable
Publishing under your own name is a security decision before it is a career one
Building a public record under a real name creates reach and creates exposure, and the two arrive together. What actually becomes searchable, which categories of harm follow, and how to publish deliberately rather than discovering the terms afterwards.
2026-09-12 · founder-essay · 4 min read
The Default Is the Policy
What a system does when nobody chooses is what it does, and everything else is documentation
Settings pages describe what is possible. Defaults describe what happens. Since almost nobody changes a default, the default is the operative policy of a system regardless of what any document says — which makes choosing defaults the most consequential design decision most teams make without noticing.
2026-09-12 · technical-essay · 3 min read
The Face Is the Password Now
Biometrics solve the problem of remembering and create the problem of not being able to change
A face unlocks a phone, clears a border and, increasingly, confirms an identity to a service that has never met you. What biometrics genuinely fix, the property that makes them different from every other credential, and where the distinction between matching on your device and matching on somebody else's server decides everything.
2026-09-12 · technical-essay · 4 min read
The Hidden Infrastructure of Shadow AI
An unapproved tool inherits the reach of the systems it sits upstream of
Staff paste sensitive material into endpoints nobody approved. The exposure is not the pasting — it is that those endpoints sit upstream of pipelines the organisation does control, so an unsanctioned tool acquires reach it was never granted.
2026-09-12 · technical-essay · 3 min read
The Model in the Middle
When a language model sits between a person and a system, it inherits both sides' permissions and neither side's judgement
Assistants are being connected to mail, files, calendars and tools. The security properties of that arrangement are not those of a chatbot or of an integration, but a third thing: a component that acts with real authority on instructions it cannot reliably distinguish from data.
2026-09-12 · technical-essay · 5 min read
The Password Advice That Made Things Worse
Rules designed to increase entropy produced predictable behaviour instead, and the behaviour was foreseeable
Complexity requirements, forced rotation and composition rules were adopted almost universally and made outcomes worse in measurable ways. Why each backfired, what humans reliably do when given a rule they cannot satisfy honestly, and what the current guidance says instead.
2026-09-12 · technical-essay · 4 min read
The Permission You Granted Once
Consent is asked at a single moment and exercised continuously, and nothing in any interface shows you the difference
Application permissions are granted once, in a second, and then apply indefinitely to software that updates itself. The structural gap between a one-time decision and continuous execution, and the small number of practices that actually narrow it.
2026-09-12 · technical-essay · 3 min read
The Problem I Am Working On
Stated as a problem, because the solution is not mine to describe yet
Long-running work drifts away from the evidence it started with, and the drift is invisible from inside because every individual step looked reasonable. This sets out the problem I have spent my time on, what makes it hard, and why the implementation is withheld — which is a claim about intellectual property rather than a claim about the work.
2026-09-12 · founder-essay · 4 min read
The Supply Chain You Did Not Choose
Every dependency you install brings its own dependencies, and the trust decision was made by somebody else
Installing one package commonly installs hundreds. Each was chosen by the author of the one above it, updates automatically, and executes with the same privileges as your own code. What that actually exposes, why the usual advice does not scale, and the small number of measures that change the shape of the risk.
2026-09-12 · technical-essay · 4 min read
Two-Factor Authentication, Ranked
From the one that stops nearly everything to the one that is mostly a formality, with the reason for each position
The common second factors are not equivalent, and the differences are not marginal. An ordering by what each actually resists, why the weakest is still worth enabling, and the failure that defeats most of them regardless of which you chose.
2026-09-12 · technical-essay · 4 min read
What a Model Cannot Know About Itself
Asking a system to report its own confidence, its own reasoning, or its own limits produces text, and text is not evidence
Assistants are routinely asked how sure they are, why they answered as they did, and whether they can do a thing. Each answer is generated by the same process that produced the original output, which means it is a plausible continuation rather than an observation. What this rules out, and what to measure instead.
2026-09-12 · technical-essay · 4 min read
What a Security Audit Does Not Cover
The scope is the whole content of the finding, and it is the part left off the badge
An audit report says something precise about a defined system at a defined moment. The badge on the website says something vague about a company forever. How to read the difference, and the four questions that recover the real meaning from a claim of having been audited.
2026-09-12 · technical-essay · 3 min read
What a VPN Does Not Do
The three things it genuinely changes, and the many it is sold as changing
A VPN moves the point at which your traffic joins the public internet. That is a real and sometimes valuable change. It is not anonymity, it is not encryption of things that were not already encrypted, and it does not remove you from the systems that identify you.
2026-09-12 · technical-essay · 4 min read
What Happens When the Company Dies
Acquisition, insolvency and discontinuation are data events, and almost nobody plans for them
Assessments of a service ask whether it is secure and whether it is trustworthy. They rarely ask what becomes of the data if the company is sold, wound up, or simply stops. Those three endings have different mechanics and different consequences, and all three are ordinary.
2026-09-12 · technical-essay · 4 min read
What I Look For When I Audit a Service
A short list, in the order I check it, and why the first item is almost always the last word
A practical procedure for assessing whether a service can do what it says about your data, written as an order of operations rather than a checklist. Most assessments are settled by the first two questions; the rest establish how much of the answer was deliberate.
2026-09-12 · founder-essay · 4 min read
What Smart Devices Send Home
A speaker that only listens for its wake word still has a great deal to say about you
The argument about whether these devices record everything misses the more consequential point: the traffic they generate when working exactly as described already describes your household in detail. What leaves, what it reveals, and the questions that separate a device you can live with from one you cannot.
2026-09-12 · technical-essay · 4 min read
What Your Browser Extension Can See
The most privileged software on your machine is usually the software you thought about least
An extension with permission to read and change data on all sites can see everything the browser sees: the pages, the forms, the session that keeps you logged in. Why that permission is so commonly granted, what it actually permits, and how the risk arrives long after installation.
2026-09-12 · technical-essay · 3 min read
Who Are You Actually Defending Against
Most personal security advice is given without asking the one question that determines whether any of it applies
Advice is dispensed as though everyone faced the same adversary. The measures that protect against an opportunist are different from those that matter against somebody who knows you, and different again from an adversary with legal authority. A method for working out which set applies to you, and why copying somebody else's precautions usually wastes effort.
2026-09-12 · technical-essay · 4 min read
Who Holds the Key
One question organises almost everything in practical security, and most products answer it without being asked
Transport encryption, storage encryption, password managers, backups, messaging, biometrics and cloud storage look like separate subjects. They are one question repeated: who is in a position to read this, and what would it take. A single frame for the whole field, and where each common arrangement sits inside it.
2026-09-12 · technical-essay · 6 min read
Why I Publish What My Software Cannot Do
Stated limitations are expensive, which is exactly why they are worth reading and worth writing
Every product page lists capabilities. Almost none list the things the system is structurally unable to do. An argument for publishing limitations as a design discipline rather than a confession, and what changes in the engineering when you commit to it.
2026-09-12 · founder-essay · 4 min read
Why Phishing Still Works
It is not that people are careless. It is that the message arrives when the story already makes sense.
Phishing is usually explained as a failure of user attention, which is why twenty years of telling people to be careful has not fixed it. A better explanation is that a convincing message arrives at a moment when it fits, and fitting is cheap to arrange.
2026-09-12 · technical-essay · 3 min read
You Cannot Verify the Code You Are Running
Open source proves what was published, not what was executed, and the gap between those is where trust actually sits
Published source is treated as evidence that a program does what it claims. It is evidence about a repository. Between that repository and the binary on your machine sit a compiler, a build machine, a distribution channel and an update mechanism, none of which the source describes. What closes parts of the gap, and what remains open.
2026-09-12 · technical-essay · 4 min read
Your Password Manager Is Not the Weak Link
The objection is intuitive, common, and wrong for a reason worth understanding
Putting every password in one place sounds like concentrating risk, and the intuition is not stupid. It is wrong because it compares the wrong two options: not a manager against perfect discipline, but a manager against what people actually do instead.
2026-09-12 · technical-essay · 3 min read
Browser-Level Encryption: What the Padlock Does Not Cover
The difference between a connection nobody can read and a file nobody can read
TLS protects a file while it moves. It stops protecting it the moment it arrives. This is the distinction between transport encryption, encryption at rest, and end-to-end encryption performed in the browser — stated in terms of who holds the key at each stage, because that is the only question that separates them.
2026-09-11 · technical-essay · 4 min read
Free Is a Price
What a service costs when it does not charge, and how to work out what you are paying before you sign up
Every service is paid for. When the user is not the payer, the revenue has to come from somewhere, and the shape of that somewhere determines what the product is motivated to do. A method for reading a business model off a service before you build anything on top of it.
2026-09-11 · technical-essay · 4 min read
What a Breach Actually Leaks
Why the answer to 'were passwords exposed?' matters less than people think, and what to look at instead
Breach notifications are written to be survivable, which makes them poor instructions. This separates the categories of data a breach can expose by how long each stays dangerous, and gives the reader a way to decide what to do that does not depend on the wording of the notice.
2026-09-11 · technical-essay · 4 min read
What the Model Remembers
Where the text you paste into an assistant goes, who can read it, and which of the usual reassurances actually mean something
Pasting a document into a chat assistant is not the same kind of act as searching for something. This separates what happens to that text — the request, the retention, the human review, the training set — and gives the questions that distinguish a service that cannot read your input from one that merely says it will not.
2026-09-11 · technical-essay · 5 min read
Who Holds the Key: How files.vualet Differs From Dropbox by Design
A comparison written while one side of it is still being built, and labelled as such
Dropbox encrypts files in transit and at rest, and holds the keys that open them. files.vualet is being built so that it cannot. This sets out the architectural difference, separates what is true today from what is intended, and gives the reader the three questions that settle it for any service.
2026-09-11 · technical-essay · 4 min read