☄️ Anthropic revealed how hackers from around the world tried to turn Claude into a weapon Anthropic has published the most detailed report to date on attempts to abuse Claude - over the past eight months, the company has identified and stopped dozens of operations where they tried to use AI for harm: from cyber attacks and surveillance to weapons development. ✏️Each case was stopped, and the lessons learned were used to strengthen protection; where appropriate, Anthropic has shared data with authorities and other AI companies. The company emphasizes that these are not typical cases, but the most sophisticated attempts at abuse that have been recorded - they are publishing them specifically to show where the AI threat industry is heading, where protection works and where it needs to be strengthened, and so that other platforms can recognize similar activity on their own. What's inside the report: 🔴 Cyber attacks - Russian state hacker group automated espionage through AI agents; financially motivated hackers (including those associated with ShinyHunters) hacked companies en masse and extorted money; Chinese students conducted a full-fledged espionage operation with the help of AI agents who themselves looked for zero-day vulnerabilities. The theft of AI access as a separate business - criminals hunt for access keys to AI models, resell them and use them instead of their own computing power; one actor spent a month unsuccessfully trying to gain access to a yet-to-be-released Claude model.🔴 Information operations—Russian, Iranian, and other state propaganda organizations have used Claude to create fake news networks, thousands of fake accounts, and even clone the voice and text style of a real activist to infiltrate his contacts inside Iran.🔴 Surveillance—from commercial spy platforms profiling users in Iran and the Persian Gulf, to the Mali government system intercepting data from 25 million SIM cards, and Chinese surveillance operations on Uighurs, Tibetans and Falun Gong practitioners.🔴 Conventional weapons development - Claude was used to write software for missile guidance systems in Yemen (including a real field test), developing a swarm of kamikaze drones in Russia, and creating software for electronic warfare and air defense suppression in China.🔴 Biological risks - researchers tried to bypass defenses to get help in working with the chikungunya virus, bird flu and orthopoxviruses - in some cases under the guise of civilian grants, but with military institutional links.🔴 Fraud - a Chinese studio created a network of 20+ fake dating applications, where thousands of Claude AI-persons corresponded with real people, posing as real people.🔴 Illegal distillation of models - Alibaba, Moonshot (Kimi), DeepSeek, Zhipu and Xiaomi stole Claude's reasoning data en masse through networks of thousands of fake accounts to train their own models - with some companies secretly replacing their AI's responses with Claude's without alerting users.#news#Anthropic📷 BlueScreen | Digital Ka