Anthropic Reveals How AI Is Being Misused by Bad Actors
Anthropic has published its most detailed threat intelligence report yet. It documents how bad actors have tried to use Claude for cyberattacks, surveillance, influence operations, weapons development, and AI model distillation, and how the company found and stopped them.
- A suspected Russia-linked group used Claude across phishing, network intrusion, data theft, and malware development while targeting more than 20 organizations.
- Hackers linked to ShinyHunters used AI agents to scan 1.8 million Android apps for exposed secrets and help run breaches across multiple companies.
- Claude was also being used for government surveillance and across six weapons programs, including guided rockets, ballistic missiles, a hypersonic glide vehicle, and autonomous FPV drones.
- Influence operations included fake news networks, political accounts, propaganda campaigns, and systems designed to imitate the writing styles of real people.
- One dating operation reportedly created 4,700+ AI personas that interacted with more than 25,000 people across over 20 dating apps, generating around 2.36 million Claude messages in two weeks.
- Anthropic also says Chinese AI companies, including Alibaba, Moonshot, DeepSeek, Z AI, Xiaomi, SenseTime, and MiniMax, used Claude to help improve their own models through distillation. Alibaba alone allegedly generated more than 151 million Claude exchanges between May and July.
Perhaps the most striking claim is that the company says Moonshot and DeepSeek were sometimes routing requests from their own users through Claude without telling them, and then collected the exchanges for model training, while other labs were actively trying to extract Claude’s hidden reasoning. You can read the full report here.
Meta Changes AI Suggestions After Invasive Personal Questions
Meta says it has changed how its AI chatbot suggests prompts after a viral video showed the assistant asking a user intrusive questions about her children and family.
- The issue surfaced after Instagram user Kalie Robins shared a video showing Meta AI suggesting the question “Who is the child passenger?” beneath a video of her and her child.
- Robins said the chatbot then connected information about her daughters from her own posts and those shared by relatives, while suggesting further questions about their ages and where her family lives.
- She also claimed Meta AI surfaced photos of her daughters that had previously been posted to the platform, including an image she said she had deleted years earlier.
- Meta acknowledged that the feature “missed the mark” and said it should not have generated prompts asking about such personal topics.
The company says the feature is designed to help users discover more information about posts they can already access, and that it has now fixed the issue responsible for generating prompts about sensitive personal subjects.
Turn any Google Sheets into an app
- Open the Google Sheet you want to use.
- Click the Gemini icon in the top right.
- In the settings, select “Create canvas.”
- Describe the app you want to create. For example: “Create a dashboard based on this data.”
- Gemini will generate the app, with the sheet essentially serving as its database.
You’ll not only be able to visualize the data but also input new data.
OpenAI launches ChatGPT for financial services
ChatGPT for Financial Services is a tailored ChatGPT Work experience that combines built in financial data with GPT 6 Astra’s reasoning. It helps financial teams conduct research, build financial models and create customized client materials.
Andrew Tulloch leaves Meta for Anthropic after 11 months
The prominent AI researcher Andrew Tulloch is leaving Meta to join Anthropic after less than a year at the company’s Superintelligence Labs. He’ll now work on Anthropic’s inference and performance team.
OpenAI pauses Pro subscriptions amid Astra demand
The company has temporarily stopped new sign ups for its $200 per month Pro plan, saying the tier puts the most strain on its infrastructure amid unprecedented demand for Astra.
Meta’s Muse reaches No. 2 on the US App Store
Meta’s AI agent Muse has surpassed 83,000 US iOS downloads, pushing it to No. 2 on the App Store’s Top Charts just days after launch.
OpenAI adds AI safety official to its board
The company has appointed AI safety researcher Paul Christiano to the OpenAI Foundation board, where he will join its Safety and Security Committee. Christiano has warned that rapid AI progress could lead to a loss of human control and says the industry, including OpenAI, is not yet doing enough to reduce that risk.