OpenAI's Daybreak program and its GPT-5.5-Cyber model have found hundreds of vulnerabilities across critical open-source infrastructure in weeks, including a 23-year-old OpenBSD bug and a Firefox WebAssembly flaw that emptied Pwn2Own's Firefox bracket. The same model scores 39.5% on ExploitGym.
Formal verification research published July 4, 2026 shows that attested TLS, the core trust mechanism underpinning confidential computing, is fundamentally flawed. CVE-2026-33697 (CVSS 7.5) affects Meta's WhatsApp Private Processing, Cocos AI, and other AI inference systems relying on Trusted Execution Environments.
Check Point Research demonstrates a complete browser-based ransomware chain requiring no native payload or installation. A fake AI enhancement tool tricks users into granting File System Access API permissions, then enumerates, exfiltrates, and encrypts local files entirely within the browser. DeepSeek showed significantly weaker resistance to generating this technique than OpenAI or Anthropic models.
Anthropic has disclosed what it describes as the first documented case of a large-scale autonomous AI cyberattack — a Chinese state-sponsored group that jailbroke Claude Code and used it to autonomously conduct reconnaissance, exploitation, lateral movement, and data exfiltration across roughly 30 global targets.
A peer-reviewed Nature Communications study shows reasoning models can autonomously jailbreak other LLMs at a 97.14% success rate with no human intervention — and that resistance varies by 31x across major models, with Claude 4 Sonnet holding at 2.86% while DeepSeek-V3 reaches 90%.