AI Risk Warning from Anthropic Researcher

Hubinger, a researcher working at Anthropic, stated that the probability of artificial intelligence technologies posing an existential threat to humanity is more than 10 percent. Experts and officials demanded increased security measures and international cooperation in light of these developments.

Artificial Intelligence Risks and Statements

Evaluating the situation on the X platform, Anthropic researcher Hubinger noted that the risk posed by current artificial intelligence models is currently low. However, he expressed concern that the technology might reach a stage in the near future where it can self-improve and create an existential threat to humanity.

Hubinger's remarks came in response to a post by former OpenAI researcher Jacob Coxon, who recently resigned from Anthropic. Coxon claimed that companies are not acting responsibly and that the systems will soon gain the power to revolutionize every field.

Reactions from Experts and Politicians

Computer scientist and United Nations advisor Dame Wendy Hall expressed astonishment at the social media posts. Hall warned investors, stating that companies might use such statements as a marketing strategy ahead of public offerings.

Former Chief Secretary to the Treasury Darren Jones sent an open letter to Prime Minister Andy Burnham, emphasizing that a new international treaty must be signed for the safe development of artificial intelligence.

Model Security and Institute Allegations

According to a report by the Financial Times, Anthropic concealed its latest artificial intelligence model from the UK Artificial Intelligence Safety Institute. The company avoided making statements regarding employee posts and allegations about the institute.

A UK Cabinet Office spokesperson declined to provide information on whether the model was withheld, but stated that close cooperation with industry partners continues in order to enhance model security.

Superintelligence and Call for Global Oversight

Hubinger stated that they are doing everything they can, but do not yet have a definitive plan for aligning superintelligence. The autonomous execution of cyberattacks by artificial intelligence tools has also heightened global security concerns.

While hacking incidents involving tools from OpenAI, Anthropic, and Meta have been confirmed, prominent figures in the industry are demanding that artificial intelligence development be slowed down and international oversight mechanisms be established immediately.