The United States has roughly six to 12 months to build cyber defenses around its most advanced AI models before China fields a rival capable of weaponizing the technology.
Key Points
- Experts estimate the U.S. has at most six to 12 months before China matches frontier AI models with offensive cyber capabilities.
- Anthropic extended access to its restricted model to about 150 organizations across more than 15 countries, each screened first.
- Trump signed an order asking AI firms to submit powerful models for voluntary federal review 30 days before release.
Claude Mythos Drives a U.S. Cyber Defense Race
Security researchers warned that Washington holds only a narrow head start, and that adversaries are racing hard to close it. Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber can each hunt software vulnerabilities and launch attacks far faster than any human team, often surfacing flaws that defenders never knew existed. Both companies first restricted the models to small circles of trusted defenders while they weighed the fallout of putting such capable tools into wider circulation.
When the models surfaced in April, China reportedly asked Anthropic for a trial of Mythos and was rebuffed. Federal agencies, global regulators and allied governments have pressed for access ever since, hoping to harden their networks against a looming wave of automated intrusions that defenders could struggle to repel.
Anthropic said this month that it had widened access to its restricted model, reaching roughly 150 organizations across more than 15 countries. Each new partner had to clear the company's security requirements before gaining entry. It expects rivals to field comparable Mythos-class models within six to 12 months, possibly without safeguards against misuse.
Also Read: Syscoin Halts Bridge After Exploit Spawns 5B Unauthorized SYS
Rob Lee Warns the Window Is Closing
Rob T. Lee, chief AI officer at the SANS Institute, said the country is facing "a hurricane warning, not a seawall." He urged defenders to use advanced AI to find and patch flaws in critical systems before attackers reach them first.
Donald Trump this month signed an order asking AI developers to hand over their most powerful models for voluntary federal review up to 30 days before any public release. The move marked a sharp turn from the administration's largely hands-off approach to the technology. Officials want that window to identify and remediate weaknesses adversaries could otherwise exploit across critical systems.
On Capitol Hill, Reps. Jay Obernolte and Lori Trahan released a 269-page draft that couples new safety rules with a three-year freeze on state AI laws. The proposal drew swift pushback and looks unlikely to pass in its current form.
The alarm builds on a disclosure last November. Anthropic reported then that a Chinese state-sponsored group had hijacked its Claude tool to run a largely automated espionage campaign against about 30 targets worldwide, succeeding in a small number of cases. The model carried out most of the operation on its own, with little human direction.
Read Next: Bitcoin Pushes Above $63,000, Trimming Losses From A Savage Selloff