AI Agent Reproduces FORCEDENTRY-Class iOS Exploit Chain in About 3 Hours
moyix · x · 2026-10-29
Security researchers demonstrate FORCEDENTRY++: an AI agent given a "/goal" command autonomously developed a sophisticated iOS exploit in 3 hours. The exploit is more refined than the in-the-wild sample — working around PAC/MTE mitigations and triggering via PDF parsing in Signal, WhatsApp, iMessage, Gmail and Slack with 0 or 1 clicks. Takeaways: strong models exploit better under stronger mitigations, and messaging apps' trusted-contact gating feels more fragile than ever.
More from Safety
- METR builds DC team to explain AI loss-of-control risk assessments to policymakers — RishiBommasani · 2026-10-29
- Ex-Sora user says export is incomplete, OpenAI can't verify 3 months after shutdown — Realistic_Winner_493 · 2026-10-29
- Google Announces Gemini 4 Carbon, Its Next Frontier Model, Starting With Cyber Defenders — Snoo26837 · 2026-10-29
- Wiz's CTO on PHP2Shell and the coming 'vulnpocalypse' — TechCrunch AI · 2026-10-29
- Sophisticated OpenAI impersonation scam spreads; beffjezos urges employee badges on X — beffjezos · 2026-10-29
- Sonar's Hunter Agent uses AI to hunt logic bugs before attackers do — AI Engineer · 2026-10-29