Anthropic added military and disinformation cases to its report on the misuse of Claude models
Anthropic expanded its safety report on the misuse of Claude models with six cases of military use – including an attempt by Russia to develop a swarm of combat drones – and nine disinformation campaigns from Russia, Iran and Turkey. Previously, the report mainly described attempts at misuse for biological weapons research.
Anthropic added six cases of military use to its safety report on the misuse of Claude models – three in China, two in Russia and one in Yemen. These include an attempt by Russia to develop an autonomous swarm of military drones, which the company says it managed to detect and thwart. The report also describes nine disinformation operations run from Russia, Iran, Turkey and other countries, targeting audiences on six continents; according to the company, the actors used Claude models not only to generate individual texts, but also to plan entire campaigns and manage fake profiles and disinformation websites. According to the report, the content also appeared in media outlets linked to the Russian government, such as Sputnik or RT.
The 154-page report previously focused mainly on attempts to misuse Claude models for biological weapons research. The company described case studies in which users – from Russia, China and Iran, according to sources – tried to bypass safety filters and obtain information that could be used to develop biological weapons; in one case, a researcher from an “unsupported region” spent several weeks planning experiments involving avian influenza, but the system restricted their access to only the weakest Claude models. Sources differ on the number of recorded biological cases – some report five, others six.
According to the report, the company stressed that it cannot confirm with certainty an intent to cause harm in the examples mentioned, because the same information can also be used to develop vaccines. The accounts in question were blocked, but Anthropic did not disclose the names of the research institutions or the specific countries where the incidents occurred.
Why it matters
The findings show that commercially available AI models are becoming targets of attempts by state and non-state actors to use them in military, biological and disinformation contexts, forcing providers to strengthen misuse detection and companies and regulators to take this risk into account when deploying and overseeing AI systems.
What was added since the original report
Verified updates
-
Attempt by Russia to develop autonomous swarms of military drones; 6 military cases: 3 in China, 2 in Russia, 1 in Yemen; 9 disinformation operations run from Russia, Iran, Turkey across 6 continents; Creating fake profiles and propaganda websites as a specific misuse tactic; Expansion of the geographical scope of the cases to five specific countries
- Attempt by Russia to develop autonomous swarms of military drones
- 6 military cases: 3 in China, 2 in Russia, 1 in Yemen
- 9 disinformation operations run from Russia, Iran, Turkey across 6 continents
- Creating fake profiles and propaganda websites as a specific misuse tactic
- Expansion of the geographical scope of the cases to five specific countries
-
Specific focus on avian influenza research as an example; Users from specific countries: Russia, China, Iran; Extended planning of experiments ('spent weeks planning'); Safety filters restricted access to the weakest Claude models; Discrepancy in the counts: 5 vs 6 cases (the new article focuses on specific cases of biological research)
- Specific focus on avian influenza research as an example
- Users from specific countries: Russia, China, Iran
- Extended planning of experiments ('spent weeks planning')
- Safety filters restricted access to the weakest Claude models
- Discrepancy in the counts: 5 vs 6 cases (the new article focuses on specific cases of biological research)
-
The report from Anthropic contains details of attempts to misuse Claude to develop biological weapons; The report is 154 pages long; 6 specific cases of misuse in biosafety identified; Attempts involving biological weapons were thwarted and detected
- The report from Anthropic contains details of attempts to misuse Claude to develop biological weapons
- The report is 154 pages long
- 6 specific cases of misuse in biosafety identified
- Attempts involving biological weapons were thwarted and detected
Two audiences, two different impacts
What this means
For individuals
The report shows that AI model providers monitor and disclose specific cases of misuse, including account bans, which is useful to know when assessing the trustworthiness and limitations of commercial AI services.
For a business
Companies deploying Claude models or similar AI systems must account for the fact that the provider actively monitors misuse and that even commercially available models may be targeted by attempts at military, biological or disinformation misuse – this is relevant to vendor and compliance assessments of AI tools.
Risks and complianceCheck the original
Event sources
confirmed by 3 independent sources · 4 publishers, 3 independent. We count feeds from the same owner only once.