- Anthropic is providing select technology companies with early access to an unreleased AI model called Mythos, according to a Bloomberg Tech report published April 8, 2026.
- The early access program is framed around cybersecurity preparedness, aimed at helping firms anticipate and defend against AI-enabled attacks.
- Theresa Payton, CEO of Fortalice Solutions and former White House Chief Information Officer under President George W. Bush, appeared on Bloomberg Tech to discuss the defensive security implications of the program.
- The move comes as AI capabilities expand rapidly and security researchers warn that advanced models may lower the barrier to sophisticated cyberattacks.
What Happened
Anthropic has granted a cohort of technology companies early access to Mythos, a powerful AI model that has not yet been released to the public, according to Bloomberg Tech, which reported the story on April 8, 2026. The company’s stated rationale is to allow participating firms to stress-test their defenses and prepare for the category of cyberattacks that could become more feasible as AI systems grow more capable. Anthropic has not publicly disclosed which companies received access or the criteria used to select them.
Why It Matters
The decision to distribute an unreleased model for security preparation purposes reflects a broader tension in the AI industry between advancing capabilities and managing the offensive potential those capabilities create. Anthropic has previously outlined its approach to pre-release risk evaluation through its Responsible Scaling Policy, which requires internal safety evaluations before deploying models above certain capability thresholds. Offering Mythos to external partners before public release represents an extension of that framework into direct industry collaboration.
Theresa Payton, CEO of Fortalice Solutions and former White House Chief Information Officer during the George W. Bush administration, discussed the program’s significance on Bloomberg Tech alongside hosts Caroline Hyde and Ed Ludlow. Payton has previously argued that organizations need adversarial AI simulations to adequately prepare their security postures, a position consistent with the framing Anthropic appears to have adopted for the Mythos program.
Technical Details
Bloomberg’s report describes Mythos as a “powerful” model, though the company has not published a technical report, model card, or benchmark comparisons as of April 9, 2026. The specific capabilities that make Mythos relevant to cybersecurity preparation — whether in areas such as code generation, vulnerability identification, social engineering simulation, or another domain — were not detailed in publicly available information at the time of publication. The primary source is behind a Bloomberg subscription paywall, limiting independent verification of technical claims beyond the summary distributed through Bloomberg’s RSS feed.
What is confirmed: the program is explicitly scoped around defensive security use cases, and early access is being provided ahead of any general availability announcement.
Who’s Affected
The immediate cohort consists of the unnamed technology companies that received early access. If the program surfaces novel attack vectors or defensive techniques, findings could propagate to enterprise security teams, critical infrastructure operators, and cloud platform providers that rely on AI-integrated systems. Cybersecurity vendors building AI-based detection and response products are also a likely audience for any published outcomes from the program.
More broadly, the program sets a precedent for how frontier AI labs might coordinate pre-release model access with industry partners under a security rationale — a framework that regulators and standards bodies in the United States and European Union have been developing guidance around.
What’s Next
Anthropic has not announced a public release timeline for Mythos, nor has it indicated whether findings from the early access program will be published. The company’s next scheduled external communications around model releases and safety evaluations would typically follow internal review completion. Given Anthropic’s past cadence — Claude 3 family models were released throughout 2024, and subsequent generations followed in 2025 — a broader announcement regarding Mythos could come within weeks of the early access program concluding, though no date has been confirmed.
