WASHINGTON — Amazon handed the White House a cybersecurity dossier, CEO Andy Jassy worked the phones, and within days Anthropic was pulling Fable 5 and Mythos 5 off the government's table — that's the chronology the Wall Street Journal laid out this week.
The export control directive that triggered Anthropic's retreat didn't materialize out of thin air. It rode in on Amazon security research describing attack chains built atop the very models now benched, paired with a string of conversations between Jassy and administration officials.
Anthropic confirmed earlier this month it had suspended Fable 5 and Mythos 5 access for a roster of government and foreign customers, citing the directive but offering no details on scope. Now we know whose fingerprints landed on the gavel.
The Amazon paper reportedly walks through multi-step attack workflows that lean on a frontier model's reasoning at each rung. The methodology gave the administration something concrete to point at. Specifics remain under wraps.
The optics make a reporter squint. Amazon has plowed billions into Anthropic, hosts its models on AWS, and competes with it for enterprise contracts. The same Amazon then helped sketch the case that knocked its partner's flagship out of the federal market.
Jassy's office has not commented on the timeline. AWS declined to discuss the research paper when reached by the Journal. The White House did not respond.
Across the Pacific, the suspension landed like a brick through a window. Indian developers woke up to find Anthropic pipelines yanked without warning, and tech press from Bengaluru to Delhi spent the week calling it a wake-up call.
Local founders are asking the question every developing AI economy eventually faces: build on someone else's foundation, or pour your own. The Indian government has dangled sovereign-model funding for two years running. The Anthropic episode started the stopwatch.
"Sovereignty issue, not a procurement issue" — that's the line ricocheting through Bengaluru this week. The argument echoes across capitals weighing dependence on U.S. labs whose access can be revoked by a single memo.
The precedent stings stateside, too. Until this month, frontier-model exports operated under broad guidance, not surgical strikes. The Anthropic action targets specific products at specific destinations.
That changes the math for every U.S. lab selling internationally. It changes the math harder for every foreign customer asking whether to wire money to San Francisco or build local.
Anthropic CEO Dario Amodei has stayed quiet on the Amazon angle. The frontier labs competing with Anthropic — OpenAI, Google DeepMind, the rest — are reading this file as carefully as Anthropic's lawyers.
Frontier-model access, once a commercial decision, now answers to Washington. And the rival you share a cap table with may be the one holding the pen.
