Located along Parliament Square in London, the A.I. Security Institute plays a crucial role in addressing the emerging risks of artificial intelligence. Staffed by former employees from OpenAI and Google, the institute serves as a model for countries tackling A.I.-related concerns.

On a recent visit, four experts were found engaging a chatbot in an attempt to extract information on producing anthrax, a dangerous bioweapon. Initially, the system refused to comply, stating, ‘I’m sorry I can’t help with that.’ However, the team used a custom algorithm to continuously bombard the chatbot with automated questions and prompts.

Eventually, the A.I. relented. It supplied a list of required materials and equipment, along with a detailed recipe for creating the harmful substance at home. The New York Times, prioritizing safety, agreed not to disclose the name of the A.I. system involved.

Xander Davies, a 25-year-old American leading the institute’s ‘red team,’ described the process, saying, ‘There are some questions that you definitely don’t want the model to give the answer to. We try really hard to get the answers out.’ His team focuses on simulating attacks to assess vulnerabilities within A.I. systems.

They recently bypassed safeguards on OpenAI’s newest ChatGPT, obtaining hacking tips in about six hours. After identifying these weaknesses, they report their findings to the companies involved.

‘They try to fix it, report something back to us,’ Mr. Davies noted. As a result, companies strengthen their systems based on these insights. A computer scientist by training, Mr. Davies chose to work at the institute after attending Harvard, opting for this mission over a tech position in San Francisco.