Trusted access for the next era of cyber defense

The more consequential detail buried in this announcement is the access model itself: GPT-5.4-Cyber is not a public release but a vetted-access program, meaning OpenAI is deliberately segmenting its most capable security tooling behind a credentialing layer rather than shipping it broadly.

This pairs directly with Anthropic's release of Claude Mythos Preview, covered here on April 17, which was framed as a bid to repair relations with the Trump administration over national security concerns. Both labs are now running parallel gated cybersecurity programs within days of each other, which looks less like coincidence and more like a race to establish credibility with government and enterprise security buyers before procurement cycles close. OpenAI's simultaneous $10M API grant program (covered in 'Accelerating the cyber defense ecosystem') adds a subsidy layer that Anthropic has not matched publicly, giving OpenAI a structural advantage in seeding adoption among smaller security firms that might otherwise default to whichever model is cheapest.

Watch whether Anthropic announces a comparable grant or subsidy program for Claude Mythos Preview within the next 60 days. If it does not, OpenAI's cost-offset strategy likely pulls security-focused startups into its orbit before Anthropic can compete on access terms.