Why AI Cyber Tools Are Becoming the New Vendor Battleground

Two of the world's most powerful AI companies both released tools this week that can autonomously hunt for security weaknesses in software. They disagree sharply on who should be allowed to use them.

Anthropic and OpenAI have each launched AI models capable of discovering vulnerabilities that human experts could previously only find with significant effort. Their approaches to access could not be more different. For marketing and technology leaders evaluating AI vendors, the split matters more than it might first appear.

Anthropic Locks the Door

Anthropic's new model, Claude Mythos Preview, operates inside a closed program called Project Glasswing. Participation is restricted to a small, hand-picked group of large organizations including AWS, Apple, Google, Microsoft, Nvidia, Cisco, CrowdStrike, and the Linux Foundation.

Control vs. Scale: How Anthropic and OpenAI Diverged on Cyber AI Risk

Anthropic limits cyber AI access via Project Glasswing. OpenAI scales broadly. Which risk philosophy fits your organization best?

Mission MediaMission Media

The company's reasoning is straightforward: the model is capable of autonomously finding severe security flaws, and Anthropic believes unrestricted access would not be safe. Mythos is not available to general customers. Partners get direct oversight and are expected to have established internal security processes and significant engineering resources.

The logic is tight control now, expand later. But tighter circles also mean slower feedback loops and a narrower base of defenders who benefit.

OpenAI Opens the Gate (With ID Required)

OpenAI's approach with GPT-5.4-Cyber is fundamentally different. The model sits inside an expanded program called Trusted Access for Cyber, which is open to thousands of individual security practitioners and hundreds of corporate security teams.

Access still requires verification checks. But the scale is far broader. OpenAI frames this as "democratized defense," arguing that defenders need wide access to advanced tools to keep pace with threat actors who are already experimenting with generative AI on the attack side.

The bet is that a larger pool of vetted users produces better collective outcomes than a small, curated cohort.

Jonny Scott, Head of Cyber Advisory at Phoenix Software, sees both positions as coherent but cautionary: "The AI hype is real, and the change is coming, for good and bad. But it's interesting to see these two companies taking such different views on how these solutions should be brought to market. Both approaches aim to strengthen defense, but they clearly have very different risk tolerances."

The Open-Weight Model Threat

The deeper concern for both camps is not each other. It is the proliferation of open-weight models that no company controls.

Edward Wu, founder and CEO of Dropzone AI, puts it plainly: "It's imminent that similar capabilities will become more widely accessible to actual attackers over the next 12 to 18 months as open-weight models catch up. For defenders, this means assuming much shorter patching windows, adopting an 'assume breach' mindset, and investing in automation to operate at machine speed and scale."

Research backs that timeline. MITRE's OCCULT framework evaluations show open-weight models like DeepSeek-R1 already achieving over 90% accuracy on offensive cyber knowledge tests. The controlled-access window is time-limited.

Implications for Technology and Security Leaders

Both models are currently restricted to defensive use cases: penetration testing, red-teaming, and code review. Neither company is handing offensive capabilities to the general public. But the question of who gets access, and how quickly, shapes which organizations can build a meaningful security advantage before these capabilities become commoditized.

Why AI Governance Beats Technology in the APAC AI Race

60% of enterprises building AI see no measurable value. Data integration and governance structures separate AI leaders from laggards in APAC.

Mission MediaMission Media

83% of APAC organizations already report rising AI-driven vulnerabilities, making the access divide a pressing regional concern. For marketing and technology leaders who influence vendor decisions, the choice between these two approaches is not purely technical. It is a question of risk tolerance, timeline, and whether your organization can qualify for the more restricted option in the first place.

The AI cyber arms race is real. The access strategy is now the battlefield.