OpenAI Compaction Flaw Let Researchers Bypass Hidden Reasoning Protections
OpenAI's compaction feature, which shrinks a long conversation to save space, could be turned into a way to copy a model's hidden reasoning out in plain text, security firm Xint has shown. OpenAI's reasoning models work through a problem before answering, and the service normally returns only the raw reasoning as an encrypted item the user cannot read. The method got past
Sources
Related
- Analysis: AI political groups have spent $78.5M on US federal candidates during the 2026 cycle; AI accelerationists spent $34M and safeguard-focused groups $39M (Luke Fountain/CNBC) OCT 27, 2026
- Anthropic's S-1 excludes Labs co-lead Ben Mann from its "Continuing Founders" and Founder LLC; Anthropic says "it's still fair to refer to him as a co-founder" (Zara Stone/The San Francisco Standard) OCT 27, 2026
- First ISC2 AI Security Certification Expected for Early 2027 OCT 27, 2026
- Sources: UK stealth AI startup 22one, which is building law enforcement tools, hit a nearly $400M valuation after raising funds from Sequoia, Lux, and others (Financial Times) OCT 27, 2026
- OpenClaw v2026.10.8 is available OCT 27, 2026