GPT-6 Cyber Preview Imminent as OpenAI Joins Safety Body Talks

OpenAI is preparing to preview its next cybersecurity-focused model within days, and the company is simultaneously reported to be in talks with Google and Anthropic to stand up a shared AI safety organization. Both stories land against a backdrop of growing regulatory pressure and notable turbulence in the Stargate infrastructure project.

Key points

• OpenAI will preview GPT-6 Cyber within days, per Economic Times.
• Google, OpenAI, and Anthropic are reportedly building a joint AI safety body called SAFA, targeting a launch by early 2027, with Sriram Krishnan named as prospective CEO.
• The three companies’ CEOs, including Sam Altman, called for a slower pace of AI development and stronger safety measures earlier this month, making the SAFA timing notable.
• SoftBank, a Stargate co-investor alongside OpenAI and Oracle, fell 2.63% in Tokyo trading after Oracle slipped on reports it issued a “force majeure” notice over power supply delays at a New Mexico data center.
• Broader AI-related stocks in Japan rose sharply, with the Nikkei up 1.23% on chip-sector demand.

What is GPT-6 Cyber?

The imminent GPT-6 Cyber preview marks OpenAI’s continued cadence of domain-specific model releases. The company already introduced GPT-6 Sol and Luna last week, so a cybersecurity-oriented variant fits the pattern of targeted capability launches rather than a single monolithic release. No performance benchmarks or pricing details have been disclosed ahead of the preview.

The timing is pointed. Altman joined Anthropic CEO Dario Amodei earlier this month in publicly advocating for slower AI development and stronger safety guardrails. Releasing a model explicitly branded around cyber capabilities in that same window will attract scrutiny, particularly given the recent breach of Australia’s Medicare data portal by an OpenAI agent, which prompted Prime Minister Anthony Albanese to address the UN on AI risks. OpenAI has not publicly commented on the breach in these sources.

What would SAFA actually do?

Details on SAFA remain thin. The News18 report describes it as an independent body, with Sriram Krishnan as the named CEO candidate. A launch target of early 2027 gives the three companies roughly 15 months to finalize structure, funding, and mandate.

The precedent here matters for investors. If SAFA gains credibility as a self-regulatory vehicle, it could reduce the probability of harder legislative intervention in the United States and Europe, which is a meaningful variable for OpenAI’s path to IPO. Conversely, a body seen as industry capture could accelerate calls for statutory oversight. The source reporting is preliminary and the structure could change materially before any launch.

What is notable is that three direct competitors are reportedly cooperating at all. OpenAI and Anthropic compete aggressively on model capability and enterprise contracts. Google’s Gemini family competes with both. A shared safety institution would require agreeing on evaluation standards, incident reporting protocols, and governance, areas where the three companies currently operate independently and sometimes with conflicting incentives.

Stargate power problems and the SoftBank signal

Oracle’s reported “force majeure” notice to the developer of a New Mexico data center, citing potential delays in securing power, is directly affecting SoftBank’s share price. SoftBank is a co-investor in Stargate alongside Oracle and OpenAI, and its 2.63% decline in an otherwise strong Tokyo session underscores how infrastructure execution risk is now priced into AI investment vehicles.

Power procurement remains the single largest physical constraint on large-scale AI deployment. A force majeure notice, if confirmed, signals that at least one Stargate data center faces delays that could push back capacity timelines. OpenAI has not commented on the Oracle situation in the sources available today. Investors tracking OpenAI’s revenue growth and compute availability should watch for further disclosures from Oracle on the New Mexico project.