Security leaders urge U.S. to lift export bans on Anthropic models

If you saw Anthropic abruptly yank its newly launched flagship models offline this week, you might have assumed hackers found a critical exploit. But the panic rippling through the cybersecurity world was not sparked by a rogue model—it was triggered by an abrupt White House export control order.

More than 100 cybersecurity executives, researchers, and technical leaders across the U.S. and allied nations published an open letter demanding action. Directed to Commerce Secretary Howard Lutnick and National Cyber Director Cairncross, the petition urges the administration to immediately lift export control directives on Anthropic’s Fable 5 and Mythos 5 models.

The roster of signatories includes former Facebook Chief Security Officer Alex Stamos, former Deputy U.S. CTO Ed Felten, and security leaders tied to Zoom, Sophos, and Stanford HAI.

The standoff began on June 12. According to The Wall Street Journal, an Amazon cybersecurity paper claimed researchers used specific prompts to coax Fable 5 into serving data that could assist cyberattacks. Following discussions between Amazon CEO Andy Jassy and the White House, officials swiftly barred foreign nationals from accessing the technology.

Because export controls extended to foreign nationals inside the country—including Anthropic's own foreign-born researchers—the lab was forced to shut off both models entirely for all customers worldwide.

Handcuffing defenders while attackers advance

The central argument from the cybersecurity community is simple: the government’s technical premise does not hold up.

Officials justified the sudden intervention by claiming Fable offered an unprecedented offensive uplift. Yet experts point out that the flagged Amazon research merely tested whether a model could identify vulnerable code and construct a proof-of-concept, an essential capability required for writing secure software in the first place.

Alex Stamos and other signatories emphasized that finding bugs and generating working exploits at superhuman levels has been standard industry reality since last year. These exact capabilities already exist across rival systems, including OpenAI’s GPT-5.5, Claude Opus, Sonnet, and Chinese open-weight models like Kimi 2.7. Luta Security CEO Katie Moussouris put it bluntly on social media: she reviewed the research paper, and it was not a jailbreak.

The letter warned that Chinese open-weight models lag top American frontier systems by only months, and state-backed adversaries likely possess unpublished capabilities beyond public benchmarks. Stripping the best tools away from defenders who need to audit both newly written software and decades of legacy infrastructure simply leaves critical systems exposed.

Signatories are now demanding that any future model regulation be grounded in transparent, scientific evaluations built with academic and industry input, rather than erratic federal edicts.

Why it matters

This dispute cuts directly to the core paradox of artificial intelligence in cybersecurity: code defense and offensive weaponization rely on the exact same reasoning capabilities. If Washington attempts to outlaw every AI model capable of locating vulnerabilities, legitimate software defenders will be bound hand and foot while malicious actors leverage open-source and foreign alternatives with zero restrictions.

For enterprise security teams and developers, this saga delivers a harsh lesson in platform risk. Relying entirely on a closed commercial model for automated red-teaming or security reviews leaves your pipeline vulnerable to overnight geopolitical whims. A single disputed internal memo was enough to erase a flagship model within 24 hours.

We think the White House’s hasty intervention looks far more reactionary than strategic. Sources familiar with the talks, including former Commerce Department officials, suggest lingering friction between Anthropic and the administration over military red lines may have fueled the heavy-handed response. Pulling top-tier tools from the public defensive toolkit does not eliminate the underlying risk; it simply blinds the defenders while the rest of the world races forward.