ai-agents

13 pieces

• 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

• AI Shopping Agents: The Missing Proof You Approved the Purchase

  Card protections assume a human clicked buy; delegated agents break that assumption.

  An ai shopping agent safe requires more than trust; it demands verifiable proof of intent. Current payment systems lack the granularity to distinguish between a delegated purchase and an unauthorised action, leaving consumers to guesswork when disputes arise.

  2026-09-14 · technical-essay · 8 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

• 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

• 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

• 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

• 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

• 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

• 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