THE CIRCUITRYYour one-stop source for all tech news

Home/Tech/OpenAI Releases GPT-5.5-Cyber Preview for Vetted Security Teams

VERIFIEDBy Xavier Rivera· ·1 min read

OpenAI Releases GPT-5.5-Cyber Preview for Vetted Security Teams

OpenAI has begun a limited preview of GPT-5.5-Cyber, a more permissive version of its latest model, for vetted cybersecurity teams. The rollout follows Anthropic's Claude Mythos Preview by one month and includes details on intended security workflows.

Source:CNBC Tech

Post

OpenAI Releases GPT-5.5-Cyber Preview for Vetted Security Teams

TL;DRAI · 60 sec read

OpenAI announced Thursday that GPT-5.5-Cyber, a specialized variation of its latest AI model, is now available in limited preview to vetted cybersecurity teams.

The model is not positioned as a major advance in raw cyber capability. Instead, OpenAI said it was trained to be more permissive on security-related tasks than the generally available GPT-5.5 version released late last month.

Vetted teams can use the preview for workflows including vulnerability identification and triage, patch validation, and malware analysis. The standard safeguards in GPT-5.5 would have restricted these activities, the company noted.

From The CircuitryThe Feed — live briefs across tech, all day.See what’s happening →

"GPT-5.5-Cyber lets a smaller set of partners study advanced workflows where specialized access behavior may matter," OpenAI stated in its blog post.

The move comes one month after Anthropic launched Claude Mythos Preview and restricted access to a select group of companies under its Project Glasswing cybersecurity initiative.

Anthropic CEO Dario Amodei met with senior Trump administration officials to discuss Mythos after the company had been blacklisted by the Pentagon weeks earlier. Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent also met with major U.S. bank CEOs about the model, while Vice President JD Vance and Bessent held a call with leading tech CEOs ahead of its release.

Why this mattersAI · ~100 words

Tap a lens to see what this story means for you.

Reader-supported

DonateBuy me a coffee →Follow@thecircuitry_ →Follow@thecircuitry.to →

Reader-supported · Daily Brief

Daily brief at 7 AM ET. Top tech stories, every morning. Sourced and fact-checked.

HELP US IMPROVE

Reader-supported

The Circuitry is a passion project I've always wanted to build, and I love the work behind it.

Running it costs real money. APIs, hosting, time. To keep improving the site and growing this into something useful for everyone, those costs have to be covered.

Any contribution is appreciated. If not, no pressure. Thanks for reading.

Buy me a coffee

OpenAI AI Models Cybersecurity

MORE IN TECH

Qualcomm launches Dragonfly C1000 CPU for data centers with Meta as key client

Qualcomm introduced its Dragonfly C1000 data center CPU engineered for agentic AI on Wednesday and confirmed Meta as a customer once production begins in 2028. The announcement underscores the mobile-centric chipmaker's drive into higher-growth data center segments where energy efficiency has become a decisive factor amid surging AI-agent workloads.

Google to lower Play Store fees in US, Europe, UK on June 30

Google is set to implement lower Play Store fees and external payment options in Europe, the UK, and the US starting June 30 as part of its Epic Games settlement. The changes introduce a 10 percent service fee on the first $1 million in annual earnings for small developers and expand globally through 2027.

CISA Warns Hackers Are Actively Exploiting Severe Ubiquiti Flaws

CISA placed three severe Ubiquiti UniFi OS vulnerabilities and one critical Lantronix command injection flaw into its Known Exploited Vulnerabilities catalog on June 23, 2026, after confirming active attacks. Federal agencies must apply patches or mitigations within three days under BOD 26-04, while Bishop Fox demonstrated the Ubiquiti issues can be chained for full remote code execution and released a free detection script.