Artificial intelligence is increasingly moving beyond simple chatbots and productivity tools. A new threat-intelligence report from Anthropic has highlighted how increasingly capable AI systems can also be exploited for cyberattacks, surveillance, fraud and potentially dangerous scientific research.
The findings are adding urgency to a growing debate over how AI companies should prevent powerful models from being used for harmful purposes.
1. AI Misuse Is Becoming More Sophisticated
According to Anthropic, malicious users are no longer relying on AI simply to generate isolated pieces of information. Some operations attempted to integrate Claude into larger systems designed to automate complicated tasks.
The company reported cases involving cyber operations, weapons-related research, surveillance, political targeting and scams. The development suggests that AI’s usefulness to bad actors can increase as models become better at reasoning, coding and handling complex workflows.
2. Concerns Around Biological Research
One of the most serious areas identified by Anthropic involves attempts to use AI for research that could potentially support biological weapons development.
The company described several cases involving research into pathogens, including chikungunya and highly pathogenic avian influenza, along with work involving viruses, toxins and related biological subjects.
Anthropic said it detected these attempts and took action against the accounts involved. The examples demonstrate why AI developers increasingly treat certain scientific requests as high-risk rather than ordinary research questions.
3. Cyberattacks Could Become More Automated
Cybersecurity is another major concern. Anthropic said Claude had been used in operations involving vulnerability research and systems connected to communications infrastructure.
The company’s broader findings suggest that AI can potentially handle a much larger portion of a cyber operation than traditional chatbot use would imply. Human operators may increasingly supervise AI systems rather than perform every technical step themselves.
This shift could lower the technical barrier for some attackers while also making existing cyber threats more scalable.
4. Surveillance Raises Privacy Questions
Anthropic also reported an attempted operation in which Claude was used to automate surveillance-style reports concerning groups and individuals, including activists and political figures.
The company said the activity was identified during a pilot stage and that it found no evidence that later stages of the system were used against real targets before the account was banned.
The episode nevertheless raises questions about the use of AI for monitoring people at scale and the safeguards needed when powerful language models are combined with large amounts of personal information.
5. AI Is Also Changing Online Fraud
The report highlighted another emerging problem: AI-powered deception.
Anthropic said one China-based operation created more than 20 dating applications containing thousands of AI-generated personas. The operation allegedly used more than 4,700 fake identities to interact with at least 25,000 users, combining automated accounts with human workers.
Such cases show how generative AI can make online scams more convincing by allowing fake identities to communicate continuously and at scale.
6. Why AI Companies Are Under Pressure
The latest revelations come as governments and technology companies debate how much responsibility AI developers should have for preventing dangerous uses of their systems.
US lawmakers are discussing proposals that could require advanced AI developers to take greater responsibility for managing major risks, including potential assistance with catastrophic threats.
At the same time, Anthropic CEO Dario Amodei has called for leading AI companies to slow the development of increasingly capable models long enough to strengthen safety measures, independent evaluation and international cooperation.
7. The Next AI Challenge Is Responsible Deployment
AI is likely to remain an important tool for scientific research, cybersecurity, business and everyday work. The challenge is ensuring that the same capabilities cannot easily be redirected towards harmful purposes.
Anthropic’s findings underline an important change in the AI debate: the question is no longer only what AI can do, but who is using it, how much autonomy it has and what safeguards stand between capability and misuse.
As models become more powerful, companies and governments may have to develop safety systems just as quickly. The future of AI could depend as much on responsible deployment as on the next technological breakthrough.