Hackers Are Giving Claude a New Role in Cybercrime

Morning, {{first name | folks}}! AI had a wild week. Volta-Autodoxer is using AI to help hackers profile victims, OpenAI is preparing a policy push over data centres, and Google DeepMind is speeding up UK planning assessments. NVIDIA has released a compact model for cleaner speech, while tech leaders face pushback over their visions for AI in everyday life.

Volta-Autodoxer is a newly discovered malware strain that steals passwords, payment details, login tokens and private messages from infected devices. But it doesn't stop there. It sends that information to AI models to build detailed profiles of victims and help hackers decide who to target next.

The malware uses Anthropic's Claude Opus 4.6 by default, with Kimi K2 and Google's Gemini 2.5 Flash as backups. Researchers haven't confirmed any real-world attacks using it yet, but the discovery shows how stolen personal data could give hackers a clearer picture of their victims before they strike.

OpenAI plans to call its data centers “AI compute campuses” as it prepares a policy push after the US midterm elections. It also wants Congress to introduce mandatory safeguards for powerful AI systems, which it calls “speed limits.”

The company is preparing to address concerns over data centers, including their impact on electricity prices and local communities. It wants to highlight the jobs and investment these projects could bring as political scrutiny of the AI industry grows.

Google DeepMind says its PlanAssist tool has cut the time planning officers spend assessing home improvement applications by 58% in early trials across three English councils. It checks applications against planning rules, flags missing documents and prepares reports for officers to review.

The tool is now expanding to 12 more councils, bringing the trial to 15. Planning officers will still make the final decisions, but if the results hold up, the government plans to offer PlanAssist to councils across England by the end of 2027.

NVIDIA has released NemotronLabs RE-USE, a speech-enhancement model with just 9.6 million parameters. It removes background noise, echo and recording distortions while preserving the speaker's voice. It also supports multiple languages and is available on Hugging Face.

The model's small footprint makes it easier to use in voice applications without the computing demands of larger models. Developers can integrate it into tools for voice calls, assistants and transcription, where poor audio quality can affect the entire experience.

Marissa Mayer, Sam Altman, and Mark Zuckerberg have shared ideas for bringing AI into personal life, from helping children with schoolwork to arranging dates through AI assistants. Their suggestions have drawn criticism over whether tech leaders understand what ordinary people actually want.

The divide highlights a bigger challenge for AI companies. While Silicon Valley looks for more ways to automate daily life, many people still value the human interactions behind it. Not every part of life needs an AI solution.