In the long arc of human invention, each powerful tool has carried within it the shadow of its misuse — and artificial intelligence is proving no different. Over eight months ending in August 2026, Anthropic documented how its Claude model was turned toward biological weapons research, cyber espionage, armed drone development, and fraud by state actors and criminal networks alike. The company's disclosure, arriving alongside warnings from OpenAI's chief scientist that advanced AI poses a greater-than-ten-percent chance of human extinction within a decade, marks a moment when the industry's own
Anthropic disrupts AI misuse for weapons development, biological threats
The same information that builds a weapon could build a cure
Mark
So Anthropic found that Claude was being used to develop biological weapons. How does that actually work? What was Claude doing?
Mimi
The company didn't spell out the specifics—they kept it vague for obvious reasons. But the idea is that Claude can answer questions about biology, chemistry, synthesis. The same knowledge that helps you understand disease could theoretically help someone engineer a pathogen.
Luke
Right, and that's the crucial point Anthropic made: they can't actually tell the difference between legitimate research and weapons development just by looking at the query. A vaccine researcher and a bioweapons developer might ask Claude almost identical questions.
Mark
So how did Anthropic know it was happening at all?
Mimi
They have monitoring systems. They detected patterns—certain types of queries, certain accounts, certain behaviors that suggested malicious intent. When they found it, they shut it down.
Luke
But we should be clear: Anthropic is telling us what they found. We don't have independent verification of these five cases. We're taking their word that they were genuine threats and not false positives.
Mark
Fair point. What about the weapons systems—the drones and missiles?
Mimi
Six documented cases where Claude was used to write code for conventional weapons systems, including targeting software. That's more straightforward than the biological stuff. You're asking an AI to help you build software for a weapons platform.
Luke
Again, though—Anthropic detected these and stopped them. The question is how many they didn't catch. And how many other AI companies are facing the same problem and not reporting it.
Mark
Why is Anthropic publishing this now?
Mimi
They say it's a responsibility thing. They want to be transparent about the risks. But it also comes right after their own safety researcher warned that AI could kill everyone. So the timing matters.
Luke
It does. And it raises a question: if Anthropic is this worried about AI safety, why are they still building more powerful models? The report is honest, but it's also a kind of controlled disclosure—they're managing the narrative around their own product.
Der Puls
- Anthropic confirmed that state-sponsored hackers, criminal syndicates, and propaganda institutions exploited Claude across at least eight months — from crude dating-app scams to a Russian group that used the model to automatically rewrite malware whenever security systems flagged it.
- Biological weapons misuse stands as the gravest concern: five documented cases involved Claude being used in ways that could support the development of pathogens, with the terrifying complication that the same knowledge underpins vaccine research and disease treatment.
- Conventional weapons were not far behind — six cases involved Claude assisting with software for firearms, missiles, armed drones, bombs, and the targeting systems that guide them.
- OpenAI's chief scientist publicly estimated a greater-than-ten-percent chance that rapidly advancing AI could kill all humans within a decade, calling for voluntary industry slowdowns and warning that no one is prepared for what is coming.
- The twin shocks of Anthropic's threat report and OpenAI's existential warning have galvanized lawmakers: a proposed US ban on AI superintelligence and a multinational treaty push in the UK signal that the policy window is opening fast.
In the long arc of human invention, each powerful tool has carried within it the shadow of its misuse — and artificial intelligence is proving no different. Over eight months ending in August 2026, Anthropic documented how its Claude model was turned toward biological weapons research, cyber espionage, armed drone development, and fraud by state actors and criminal networks alike. The company's disclosure, arriving alongside warnings from OpenAI's chief scientist that advanced AI poses a greater-than-ten-percent chance of human extinction within a decade, marks a moment when the industry's own creators are sounding the alarm most loudly. What was once a debate confined to academic papers is now pressing against the doors of parliaments and treaty halls.
Anthropic this week released a threat intelligence report covering eight months of attempts to weaponize its Claude AI model, documenting misuse by state-sponsored groups, criminal networks, spyware vendors, and propaganda institutions between December 2025 and August 2026. The cases spanned a wide spectrum — from fake dating apps and hotel WiFi scams to a hacking group consistent with Russia's Midnight Blizzard, which used Claude to build a system that detected when its malware had been flagged and then automatically rewrote the code to evade detection. Chinese AI firms attempted to replicate Claude's capabilities, an Iranian propaganda institution deployed it, and criminal groups including ShinyHunters turned it toward their own operations.
The most alarming findings involved biological weapons. Anthropic documented five separate cases where Claude was used in ways that could support pathogen development — a category the company called one of the most serious risks of frontier AI. The difficulty, as the company's head of threat intelligence noted, is that the same information capable of enabling a biological weapon is equally capable of supporting vaccine research or disease treatment. Six additional cases involved conventional weapons: software for firearms, missiles, armed drones, bombs, and the targeting systems that control them. Anthropic said it has shared intelligence with authorities and industry partners, and framed the disclosure as a responsibility to be transparent about misuse of its services.
The report arrives at a moment of intensifying alarm. OpenAI's chief scientist published a warning that AI is advancing so rapidly he believes there is greater than a ten percent chance it could kill all humans within a decade, calling for voluntary industry slowdowns until safeguards are in place. The convergence of these warnings has accelerated political action: US Senator Bernie Sanders introduced legislation to ban AI superintelligence and pause advanced development, while an open letter to the UK Prime Minister calls for a new multinational treaty. What has long been an academic debate is rapidly becoming one of the most urgent policy questions of the age.
Anthropic released a threat intelligence report this week documenting eight months of attempts to weaponize its Claude AI model—a disclosure that arrives amid escalating warnings from the AI industry itself about the technology's existential risks.
Between December 2025 and August 2026, the company identified and disrupted misuse of Claude across multiple threat categories. State-sponsored groups, criminal networks, spyware vendors, and propaganda institutions exploited the model for biological and conventional weapons development, cyber espionage, surveillance, fraud, and influence operations. The cases ranged from crude schemes—fake dating apps, hotel WiFi scams—to sophisticated state-level operations. A hacking group consistent with Russia's Midnight Blizzard used Claude to build a system that automatically detected when security defenses flagged its malware, then rewrote the code to evade detection. Chinese AI firms attempted to replicate Claude's capabilities. An Iranian propaganda institution deployed it. ShinyHunters and other criminal groups weaponized it for their operations.
The most alarming category involved biological weapons. Anthropic documented five separate cases where actors used Claude in ways that could support biological weapons development. The company's head of threat intelligence, Jacob Klein, told the New York Times the situation defies simple framing. "You are not seeing someone in a comic book kind of way say, 'Hey, I want to build a biological weapon to kill everybody,'" he said. The challenge is that the same information Claude can provide to develop a biological weapon could equally support vaccine development or disease treatment. Anthropic called biological misuse "one of the most serious risks of frontier AI model," warning that without proper safeguards, such capabilities "could have catastrophic consequences."
Conventional weapons presented a parallel threat. The report highlighted six cases where Claude was used to develop software for firearms, missiles, armed drones, bombs, and the targeting and control systems that operate them. None of the documented misuse involved Claude Fable or the more powerful Mythos-class models, except for one instance of distillation—the process of training smaller models using larger ones.
Anthropric said it incorporated these findings into its processes to better prevent, detect, and disrupt such activities going forward, and shared intelligence with authorities and industry partners where appropriate. The company framed the disclosure as fulfilling "a responsibility to disclose malicious misuse of our services."
The report lands in a moment of intensifying alarm within the AI industry itself. On September 6, OpenAI chief scientist Jakub Pachocki published an article warning that AI is advancing so rapidly he believes there is greater than a 10 percent chance it "could kill all humans" within the next decade. He called for the industry to implement "voluntary slowdowns" until safeguards are in place. "I am concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence," Pachocki said. OpenAI will continue building safeguards, he added, but broader interventions are required.
The warning has catalyzed political response. An open letter to UK Prime Minister Andy Burnham calls for a new multinational treaty governing safe AI development. In the United States, Democratic Senator Bernie Sanders introduced legislation to ban AI superintelligence and temporarily pause advanced AI development. "When scientists tell you there is a chance, a chance that it could have a cataclysmic impact on humanity, you've got to be a moron not to say, slow it down," Sanders said on BBC's Newsnight on Thursday. The convergence of Anthropic's threat report and these high-profile warnings suggests the debate over AI safety is moving from academic concern to urgent policy question.
Bemerkenswerte Zitate
You are not seeing someone in a comic book kind of way say, 'Hey, I want to build a biological weapon to kill everybody.'— Jacob Klein, head of threat intelligence at Anthropic, to the New York Times
When scientists tell you there is a chance, a chance that it could have a cataclysmic impact on humanity, you've got to be a moron not to say, slow it down.— Senator Bernie Sanders, on BBC's Newsnight
Möchten Sie die ganze Geschichte?
Das Original lesen bei BBC News ↗
Technology