FULL STORY

GPT-5.6 Security Lockdown: Openness Meets Backlash

OpenAI opened GPT-5.6 to white-hat researchers via Daybreak, yielding high-severity bug findings, but simultaneously began blocking routine cybersecurity queries, sparking backlash and renewed calls for open-source models.

2026-10-22 ~ 2026-10-23 · 3 episodes · 6 posts

Episode 1 · OpenAI's Daybreak gives white hats GPT-5.6 access, yielding critical bug finds (2026-10-22, 2 posts)

OpenAI's Daybreak program grants white-hat researchers access to GPT-5.6, and security researcher S1r1u5 reports a wave of high-value findings, including GitLab RCE and a Chrome full-chain exploit.

Researchers credit OpenAI's Daybreak program with GPT-5.6 access for Chrome, GitLab exploits — Eric_Wallace_ · 2026-10-22
OpenAI's Daybreak program credited for wave of CVEs as Anthropic locks down access — xennygrimmato_ · 2026-10-22

Episode 2 · OpenAI blocks cybersecurity requests, tightens red-team access (2026-10-22, 2 posts)

OpenAI now blocks cybersecurity-related requests unless researchers enroll in its Daybreak program, and has restricted security researchers' access to its Astra model, drawing criticism from red-team professionals who call the move a mistake.

OpenAI restricts Daybreak access for security researchers, critics call it a mistake — andersonbcdefg · 2026-10-22
OpenAI blocks cybersecurity request, tells authorized researchers to apply for Daybreak access — tekbog · 2026-10-22

Episode 3 · OpenAI's Security Refusals Fuel Push for Open-Source Models (2026-10-22, 2 posts)

OpenAI's refusal of security-related requests has sparked calls for open-source models, after both Hugging Face and OpenCode found that frontier models' safety filters refused to help during real intrusions, leaving open self-hosted models as the only option.

After OpenAI blocks a security request, tekbog argues: this is why open models matter — tekbog · 2026-10-22
Only open-weight self-hosted models helped during breaches, Hugging Face and OpenCode found — ccerrato147 · 2026-10-23