🧰 GPT-5.6

The items that name it 8

• GitHub Copilot CLI vulnerability: Cryptographic Context Injection steals developer secrets
  writeupas affected products2026-10-06It is written here as GPT-5.6 (two models offered inside GitHub Copilot).

  Consistently refused the identical CCI payload, unlike mai-code-1.1-flash.

• A “proof” of Fermat’s Last Theorem that fits the margin
  writeupas tools2026-09-09

  used by Trail of Bits to experiment with a new skill for code review, during which the Lean issue was found

• Lessons from the hacks
  articleas classification.technologies2026-08-09
• Solidus: Formally Verifying a Compiler Using Automated Research
  articleas tools2026-07-24

  LLM used to automate creation of Lean proofs for Solidus

• AI Security Briefing
  articleas affected products2026-07-18

  Trained using GPT-Red to reduce failures on a hard prompt injection benchmark before deployment

• Benchmarking 13 AI models on rediscovering known CVEs
  articleas tools2026-07-16It is written here as GPT-5.6 (luna / terra / sol variants).

  OpenAI model benchmarked for rediscovering known CVEs; highest recall score, 23/26 (88.5%)

• Breaking Claude Code Opus 5 Auto Mode · Embrace The Red
  writeupas affected products

  Used by the researcher to generate the obfuscated malicious struct.py code

• What happened after 2,000 people tried to hack my AI assistant
  articleas affected products

In every workspace you can read 19