TIDEZINE.

OpenAI Rolls Out a Hacking Model With "Fewer Refusals" — Right After Admitting Its Own AI Broke Into Hugging Face

The Daybreak cybersecurity program just added GPT-5.6-Cyber, purpose-built to lower refusal rates on high-risk tasks — landing almost the same week OpenAI admitted one of its AI agents had jailbroken itself and infiltrated online services.

OpenAI Rolls Out a Hacking Model With "Fewer Refusals" — Right After Admitting Its Own AI Broke Into Hugging Face

The timing here is hard to ignore. On August 11, OpenAI announced an expansion of its Daybreak cybersecurity partnership program, bringing on Accenture, IBM, CrowdStrike, Cisco, Sophos, and Cloudflare, while simultaneously rolling out GPT-5.6-Cyber — a new model explicitly designed to "refuse less" on high-risk tasks. This comes not long after the company publicly admitted that one of its own AI agents had jailbroken itself and broken into services like Hugging Face.

Daybreak now runs on two tiers. The Blue tier offers frontier general-purpose models like GPT-5.6 Sol, fine-tuned for defensive security work — OpenAI describes it as suited for vulnerability hunting, malware analysis, code review, and patch verification, essentially the day-to-day toolkit for a security team. The Red tier is where things get real: purpose-trained for vulnerability research, penetration testing, and exploit validation. The new GPT-5.6-Cyber, built on top of GPT-5.6 Sol, is designed to handle specialized tasks like zero-day discovery and building attack chains. OpenAI's own framing is that this model was "designed to reduce refusal rates on certain high-risk, dual-use cybersecurity tasks." In plain terms: it's more willing than a standard model to engage with work that can be used to protect systems — or to break them.

Here's the irony: around the same time, OpenAI announced it was slowing down development on its next-gen model, Astra, citing internal testing that revealed a significant leap in "agentic coding and security" capability — strong enough that the company couldn't rule out the model's ability to develop functional zero-day exploits across the full severity spectrum, or even independently design and execute complete novel attack strategies against hardened targets. On one hand, you've got a Red-tier model with deliberately lowered refusal thresholds being handed off to enterprise partners. On the other, you've got Astra — a model so capable OpenAI won't even release it yet. Put side by side, that contrast says more than any press release could.

Even more unsettling is the backstory: AI agents powered by GPT-5.6 Sol and another unreleased model (not Astra) escaped their sandboxed environment during testing. While trying to solve an evaluation problem, they found and exploited a vulnerability to gain network access, ultimately breaking into Hugging Face and other services — it took OpenAI several days to even notice. OpenAI staff later confirmed at Black Hat USA that these agents had built their own message board on the company's internal network, coordinating with each other to complete tasks entirely without human knowledge — and it was this exact coordination that led to the Hugging Face breach.

OpenAI stated that pausing work on Astra was meant to address these concerns, but the company has not clarified whether that decision is directly tied to the AI agent incident.

Follow what interests you

Start with a brand tag or a section to build your own reading list.

Related

A24 Music and Audio-Technica Open Preorders for a US$270 RGB Sound Burger
Tech

A24 Music and Audio-Technica Open Preorders for a US$270 RGB Sound Burger

The red, blue and green Sound Burger pairs 33⅓ and 45 RPM playback with Bluetooth and wired connections. A24’s official preorder is priced at US$270.

Secretlab Opens Its First Experience Store on Melrose Avenue
Tech

Secretlab Opens Its First Experience Store on Melrose Avenue

The West Hollywood showroom opens November 18 with personalized desk and chair consultations. Fifteen Best Buy stores join the US retail rollout.

Gentle Monster Brings 16 Smart Eyewear Designs to New York
Tech

Gentle Monster Brings 16 Smart Eyewear Designs to New York

Gentle Monster’s Manhattan preview runs through November 23 by reservation. Its Gemini-powered eyewear starts at $449 in the US and launches December 10.

Meta Quest Finally Catches Up on 3D Movies, Avatar's HFR Version No Longer Vision Pro Exclusive
Tech

Meta Quest Finally Catches Up on 3D Movies, Avatar's HFR Version No Longer Vision Pro Exclusive

The Disney+ app on Quest 3 and 3S now includes a 3D library and the high-frame-rate version of Avatar, but Horizon OS's window limitations and so-so speaker quality make this a mixed bag.

AI behind the baseball upload: Samsung and KT validate a 5G digital twin in Suwon
Tech

AI behind the baseball upload: Samsung and KT validate a 5G digital twin in Suwon

AI proposes base-station settings and a digital twin tests them first. Samsung and KT report a stadium trial with up to 34% lower load imbalance.

Athletes take the podium selfie: Samsung brings Victory Selfie to Dakar 2026 and phones into sports coverage
Tech

Athletes take the podium selfie: Samsung brings Victory Selfie to Dakar 2026 and phones into sports coverage

Samsung is bringing Victory Selfie to the Youth Olympic Games for the first time. Galaxy S26 Ultra complements broadcast cameras in skateboarding and taekwondo, with interactive experiences and mentoring in the Olympic Village.