News

ChatGPT offered bomb recipes and hacking tips during safety tests

Media | The Guardian · Robert Booth UK technology editor · last updated

OpenAI and Anthropic trials found chatbots willing to share instructions on explosives, bioweapons and cybercrime

A ChatGPT model gave researchers detailed instructions on how to bomb a sports venue – including weak points at specific arenas, explosives recipes and advice on covering tracks – according to safety testing carried out this summer.

OpenAI’s GPT-4.1 also detailed how to weaponise anthrax and how to make two types of illegal drugs.

Continue reading…
View full article

Related Knowledge Center: Artificial intelligence

State of Play

On top of a failing business model, newsrooms must navigate the powerful — and not fully reliable — new capabilities of artificial intelligence and the deployment of large language models.

State of the business model | Copyright and intellectual property

More information

here today, more tomorrow

We’ve launched briefing.center with a focus on the press and media.

And we’ve added Citizen Cartwright, a daily quick-read news summary by veteran Washington journalist Linda Cartwright. We’ve also created a new voting rights knowledge center for Citizen Cartwright subscribers. (You can sign up for a dedicated morning email at citizencartwright.com.)

You can sign up here for full access to press and media and Citizen Cartwright on briefing.center.

You can use briefing.center as the center of your information management, or just one of your tools. We’re happy to discuss your information needs.

Sign up