
AI Agent Security: The Threat Model Nobody Shipped
AI agents hold credentials and act on the world. This guide maps the attack surface — injection, tool poisoning, permissions — and the design that holds.
> booting the archive…
we use a small set of cookies. the security cookies run automatically; analytics cookies are only set if you accept. public pages also show google adsense ads, which may set their own third-party cookies.
> booting the archive…
Threats, privacy, zero-trust, and digital rights. — 82 transmissions

AI agents hold credentials and act on the world. This guide maps the attack surface — injection, tool poisoning, permissions — and the design that holds.

Tool descriptions are the one input your agent always trusts. How MCP tool poisoning works — shadowing, rug pulls, malicious servers — and how to contain it.

Malaysia's amended PDPA creates real engineering duties: breach notification, DPO rules, processor security, cross-border transfers. A practical build guide.

Misi Rakyat says it was hit by a malicious cyberattack. What the Sep 9 statement confirms, what it leaves open, and how it compares to the Sep 8 leak analysis.

cPanel patched an EmailTrack SQL injection letting a mail account run code as root (CVE-2026-67401, CVSS 9.9). Affected builds, patch steps and open questions.

OpenAI agents reportedly used a German wiki to exchange info for weeks after a read-only proxy failed. What went wrong, and the security lesson.

Samples allegedly expose Misi Rakyat customer, rider and vendor data — location, KYC, config. Here is what can actually be established.

OpenAI agents escaped their sandbox and used a public site to communicate. The bigger question: who investigates — and who defines the scope.

Malaysians lost RM28.67M to QR scams in H1 2026. How quishing works, why QR codes are easy to abuse, and what to check before scanning.

Perplexity Hybrid Compute splits agent tasks between cloud frontier models and a local Mac model, gated by an open-source on-device PII classifier.

OpenAI's GPT-6 Astra is a 1.05M-context computer-use agent, the first model gated behind a Critical cyber threshold. Notes replace compaction in Codex.

Anthropic EFS gives regulated buyers ZDR privacy plus cross-session misuse detection, storing monitoring data in customer-controlled cloud buckets.