AI: Anton Osika Advocates for a Safe Slowdown

Anton Osika supports a slowdown in AI developments when safety is not demonstrated and calls for public accountability. His message comes after Jacob Coxon's alert and as a framework from Anthropic suggests a probability of over 10% for an extreme scenario within the next ten years.

Quantified Risk and Public Disagreements on the Scale of the Threat

Evan Hubinger, head of the alignment team at Anthropic, personally estimates the probability that AI could kill all humans in the next decade to be over 10%. He clarifies that Anthropic does not yet have a plan to address alignment at the scale of superintelligence, although efforts are underway. Following Jacob Coxon's alert, a heated debate has erupted within the scientific community, policymakers, and tech industry specialists. Some current or former members of Anthropic have publicly expressed their support for certain aspects of Coxon's warning. Other commentators express a contrary opinion: Gary Marcus believes there is no credible scenario for human extinction in the near future, while Melanie Mitchell describes these concerns as extravagant.

Osika Calls for Proof of Control and Consideration of Coordinated Slowdown

Anton Osika believes that AI development should slow down when safety cannot be guaranteed. In a message posted Thursday on LinkedIn, he calls for public accountability for decisions that could affect everyone and argues that the designers of the most powerful systems must demonstrate their ability to control them. He advocates for international coordination, including slowing down advancements deemed dangerous if safety cannot be demonstrated. He concludes that ambition in AI should be measured by what it enables for humanity. Osika is the founder of Lovable, a Swedish vibe-coding startup.

Jacob Coxon's Initial Alert and Industry Reaction

Jacob Coxon, 27, claims to have worked for about three years as a researcher at OpenAI and Anthropic and left Anthropic this week. He criticizes both labs for their eagerness to achieve a self-improving superintelligence without adopting a responsible approach. He states on X that individuals involved in designing these systems genuinely believe they could lead to the extinction of humanity before the end of the decade, emphasizing that this is not a publicity stunt. Most AI leaders have remained silent in the face of these accusations, with the exception of Anton Osika, who shares his concerns.

Immediate Silence from Giants, Reminders of Previous Alerts

So far, the leaders of OpenAI and Anthropic have not publicly commented on Jacob Coxon's warning, and requests for comments made outside of business hours have gone unanswered. In the past, Sam Altman indicated in April that his company might need to collaborate with governments, international organizations, and rival AGI developers to address serious alignment, safety, or societal issues before any further advancements. In February 2025, Dario Amodei warned of global security threats related to advanced AI, ranging from catastrophic use to autonomous risks. In this context, Anton Osika voices support for a slowdown conditioned on the demonstration of safety.