Jacob Coxon, a former researcher at Anthropic, one of the leading AI safety companies, told the BBC that artificial intelligence staff members are "genuinely frightened" about humanity's future. Coxon warned there exists a strong chance AI could end humanity if the rate of development continues unchecked.
His comments reflect growing tensions within the AI industry between those building powerful systems and those tasked with safety oversight. Anthropic, founded in 2021 by former OpenAI executives including Dario Amodei and Daniela Amodei, was established specifically to develop AI systems safely and responsibly. The fact that researchers working at a company with safety as its core mission express deep concern about existential risk signals how seriously the threat is being taken inside AI labs.
Coxon's warning targets the accelerating pace of AI development across the industry. Companies like OpenAI, Google DeepMind, Meta, and smaller labs are racing to build larger models with greater capabilities. Each breakthrough in language models, reasoning systems, and multimodal AI narrows the gap between current technology and theoretical artificial general intelligence (AGI). Researchers worry this sprint leaves insufficient time for safety testing and alignment work, the process of ensuring AI systems behave according to human values.
The concern has roots in broader AI safety research. Leading figures including Sam Altman at OpenAI and Demis Hassabis at Google DeepMind have publicly acknowledged existential risks from advanced AI. Yet these same executives continue funding aggressive R&D programs. This contradiction creates internal friction. Researchers working on alignment and safety often feel their concerns are deprioritized against timelines pushing for capability leaps.
Coxon's decision to speak publicly to the BBC represents a rare breach of industry silence. Most AI researchers, bound by non-disclosure agreements and professional concerns about appearing alarmist, avoid public statements about existential risk. Speaking out carries career risk and could be perceived as undermining the companies investing in their work. That Coxon chose to do so anyway underscores the depth of anxiety among some technical staff.
The statement also reflects a growing divide between AI safety researchers and capabilities researchers. Safety teams argue for slowdowns and rigorous testing before deployment. Capabilities teams prioritize speed to market and competitive advantage. Regulatory pressure from the EU AI Act and ongoing US regulatory proposals has created urgency on both sides, but for different reasons. Regulators want oversight. Capabilities teams want to move fast. Safety teams want to move slow.
Coxon's comments arrive as the AI industry faces increasing scrutiny from governments, academic researchers, and civil society. The UN, various governments, and think tanks have launched inquiries into long-term AI risk. Yet concrete policy mechanisms to enforce slower development remain limited. No international body currently has enforcement power over AI labs. Individual companies set their own safety standards.
His words carry particular weight because Anthropic explicitly markets itself as building "constitutional AI" and has published numerous papers on alignment and safety. If researchers at such a company harbor genuine fear about extinction risk, it suggests the problem persists despite dedicated safety efforts.