Search posts, papers, and topics
Freek Van der Herten1 min readintro
This text is an abstract for an article on hardening Laravel applications against destructive AI agent behavior, referencing the "PocketOS incident" and suggesting safeguards like deny rules and credential isolation. It does not contain the actual article content, only a brief description.
1/10AI agents can turn minimal public hints about software bugs into working exploits, rendering traditional embargoes ineffective. The article cites a study where a GPT‑4 agent exploited 87% of a 15‑vulnerability benchmark from CVE descriptions and discusses faster releases and revocable capabilities as mitigations.
Sentry added zero‑config Agent Tracing for Laravel AI agents. The integration hooks into Laravel AI events and HTTP request events to create Chat spans for each LLM call, matching requests by provider URL prefix. Updating to sentry‑laravel 4.27 automatically instruments agents without any user code changes.
The Confused Deputy problem—where a privileged component is tricked into misusing its authority—has resurfaced in AI agents that process natural‑language prompts. The article shows real 2026 attacks, explains why agents are especially vulnerable, and outlines mitigations like least‑privilege scopes, capability‑based tokens, and human‑in‑the‑loop approvals.
Laravel Scalpel is a Laravel‑native scanner that looks for post‑deployment filesystem tampering – rogue PHP files, obfuscated backdoors, altered .htaccess/.user.ini, missing or mis‑configured .env, and diffs against a signed baseline. It ships with six built‑in scanners, baseline snapshot commands, fast vs strict hashing modes, CI‑friendly output (JSON, SARIF, GitHub annotations) and a ScanFinish…
The post introduces an Antithesis skill that lets AI agents perform mutation testing on distributed systems, injecting subtle bugs to check whether generated test suites can falsify safety properties. Using rqlite, the agents ran ~24 hours of tests, falsified 11 of 13 properties, and uncovered three real upstream bugs, demonstrating the technique’s practical value.
The article shows how to protect AI‑generated shell/SQL commands by parsing them with real AST parsers (tree‑sitter for Bash, Babel for JS‑embedded SQL) and enforcing structural allowlists plus command/path whitelists, instead of fragile regex filters.
