Vance, Bessent questioned tech giants on AI security before Anthropic’s Mythos release, CNBC reports
April 10 — One week before Anthropic rolled out its new Claude “Mythos” model, U.S. Vice President JD Vance and Treasury Secretary Scott Bessent pressed major tech leaders on AI model security and cyberattack preparedness, CNBC reported Friday.
According to the report, the call brought together Anthropic CEO Dario Amodei, Alphabet’s Sundar Pichai, OpenAI’s Sam Altman, Microsoft’s Satya Nadella, as well as the chiefs of Palo Alto Networks and CrowdStrike. The discussion centered on how to harden cutting-edge AI systems against exploitation and how industry and government should coordinate in the event of a significant cyber incident involving AI.
Anthropic declined to comment on the call, CNBC noted. Alphabet, OpenAI, Microsoft, Palo Alto Networks, and CrowdStrike did not immediately respond to separate requests for comment from Reuters.
A cautious rollout for Claude “Mythos”
Earlier this week, Anthropic announced a powerful new model but opted against a broad release, citing the risk that wider availability could surface or amplify undiscovered cybersecurity weaknesses. For now, access to Claude Mythos is restricted to a small circle of roughly 40 prominent technology partners, including Microsoft and Google, as the company continues additional testing and red-teaming.
Anthropic has said it remains in ongoing dialogue with the U.S. government about the model’s capabilities and risk profile—a sign of the increasingly formalized channels between AI labs and federal officials as frontier systems grow more capable and complex.
Why this matters
- Escalating stakes: As AI models become more capable, the potential for misuse—from automated vulnerability discovery to social engineering at scale—raises the bar for pre-release security testing and controlled access.
- Public–private coordination: High-level engagement from the Vice President and Treasury Secretary underscores Washington’s focus on incident response planning and cross-industry cooperation for AI-related cyber risks.
- Selective availability: Limiting Mythos to a vetted group of major partners reflects a trend toward staged deployments, allowing real-world evaluation while reducing the chance of unforeseen exposures.
Between innovation and exposure
Anthropic’s approach highlights a tension playing out across the AI sector: racing to deliver advanced capabilities while mitigating the possibility that those same capabilities could be turned against critical systems. Holding back general availability, even temporarily, suggests the company is prioritizing security audits, adversarial testing, and rule-tuning to close off potential abuse pathways before scaling access.
The reported government outreach to top AI and cybersecurity leaders points to a broader recognition that frontier models require more than typical software hardening. Defenses now increasingly include structured red-team exercises, live incident drill-downs, and clearer protocols for sharing threat intelligence between labs, cloud providers, and security vendors.
What’s next
As Mythos undergoes controlled evaluation, the industry will watch for signals on when—and under what guardrails—broader access might follow. Any timeline will likely hinge on the results of security assessments, alignment testing, and ongoing conversations with federal stakeholders. In the meantime, the pre-release engagement with Washington suggests future model launches from major labs may follow a similar playbook: targeted pilots, government briefings, and a phased path to scale.
CNBC’s report offers a rare window into how top policymakers and tech executives are recalibrating their coordination around AI security. It also reinforces that the world’s most advanced models are now entering a governance era where staged rollouts and continuous oversight are as central to the launch plan as performance benchmarks.
