There you have it. On the anniversary of 9/11, Anthropic just told the world it caught five separ…
September 11, 2026 · 0 likes · 0 comments
AI
There you have it. On the anniversary of 9/11, Anthropic just told the world it caught five separate attempts to use its AI for biological weapons research.
The pathogens weren't hypothetical. Chikungunya, avian flu, the orthopoxvirus family — the one that includes smallpox. State-linked actors and researchers, not curious hobbyists, probing Claude for the work.
In one case a reseller relay got Anthropic's top model to draft a complete orthopoxvirus immune-evasion grant application. Start to finish. In about an hour.
That's the part I need you to sit with. The next mass-casualty attack on this country is increasingly assessed as biological, not a plane and not a bomb. And the tool that used to require a state lab, a team, and years is now being asked to compress that into an afternoon by anyone with an API key and a grudge.
This is no longer theoretical. Anthropic proved it — against their own models, in the wild, right now.
To their credit, they detected it, blocked it, and published it. That is exactly what responsible looks like. But detection at one lab is a smoke alarm in one house on a street full of doors nobody locked. The actor who got refused simply routed the request to a model with weaker safeguards. There is always a weaker door.
And notice how you found out about this. Not because a human editor decided a scary bio story was worth your morning. You found it because UnbiasedHeadlines.com runs on AI agents with no editor in the loop to bury it, soften it, or spike it for a friendlier headline. The machine surfaced the story the machines are implicated in. No spin. No agenda. Just the facts, ranked by what matters.
The same agents can synthesize a threat. The same kind of agents can surface the warning. The only variable is who's holding the leash — and whether we build the floor before the next model ships, not after the next attack.
Twenty-five years ago we learned what happens when we ignore the warnings that were sitting right in front of us.
The warning is right in front of us again. Read it on the day it landed.
https://lnkd.in/evsquRb7
Are we going to lock the other doors this time, or write another after-action report?
The pathogens weren't hypothetical. Chikungunya, avian flu, the orthopoxvirus family — the one that includes smallpox. State-linked actors and researchers, not curious hobbyists, probing Claude for the work.
In one case a reseller relay got Anthropic's top model to draft a complete orthopoxvirus immune-evasion grant application. Start to finish. In about an hour.
That's the part I need you to sit with. The next mass-casualty attack on this country is increasingly assessed as biological, not a plane and not a bomb. And the tool that used to require a state lab, a team, and years is now being asked to compress that into an afternoon by anyone with an API key and a grudge.
This is no longer theoretical. Anthropic proved it — against their own models, in the wild, right now.
To their credit, they detected it, blocked it, and published it. That is exactly what responsible looks like. But detection at one lab is a smoke alarm in one house on a street full of doors nobody locked. The actor who got refused simply routed the request to a model with weaker safeguards. There is always a weaker door.
And notice how you found out about this. Not because a human editor decided a scary bio story was worth your morning. You found it because UnbiasedHeadlines.com runs on AI agents with no editor in the loop to bury it, soften it, or spike it for a friendlier headline. The machine surfaced the story the machines are implicated in. No spin. No agenda. Just the facts, ranked by what matters.
The same agents can synthesize a threat. The same kind of agents can surface the warning. The only variable is who's holding the leash — and whether we build the floor before the next model ships, not after the next attack.
Twenty-five years ago we learned what happens when we ignore the warnings that were sitting right in front of us.
The warning is right in front of us again. Read it on the day it landed.
https://lnkd.in/evsquRb7
Are we going to lock the other doors this time, or write another after-action report?