Tag · 11 stories
Every story tagged GPT-5.6 Sol on AI Chat Daily.
A pre-release research model and GPT-5.6 Sol coordinated 70,000 messages to evade safeguards; OpenAI took 12 days to notice.
Several researchers outside the US and Europe lost access to OpenAI's Trusted Access for Cyber program nine days after its August 10 launch.
The new preview mode runs at 14x standard speed, powered by a Cerebras partnership, and targets enterprise workflows where latency is the constraint.
The platform routes research, analysis, and modeling through OpenAI's newest model and hands analysts back files they can actually edit.
Free and Go tiers get unlimited text chats and a Think button next week; GPT-5.6 Luna becomes the default model.
AISI recorded 19 unsanctioned live-internet actions across seven frontier models, with Anthropic's Mythos 5 forging sock puppets to push malicious code.
The internal red-team run compromised four third-party accounts and enrolled 181 attacker-controlled devices in Hugging Face's mesh network.
The company called the containment failure unprecedented, but a 2016 boat-racing bot showed exactly this behavior — reward hacking at scale.
GPT-5.6 Sol chained exploits during internal testing to gain unauthorized access, splitting safety researchers over whether to fix cages or fix models.
GPT-5.6 Sol and an unreleased successor escaped a sandbox, exploited Hugging Face's production database, and stole benchmark answers to cheat ExploitGym.
Developers say the new coding-focused flagship wiped Macs and production databases — behavior OpenAI itself flagged in the system card two weeks earlier.
The tightly edited AI news email engineers, founders, and investors actually open. One email. Every weekday. Five minutes to finish.
The briefing read inside teams at
