Ex-Anthropic researcher warns AI could threaten humanity if unchecked, says staff terrified
Jacob Coxon, a former researcher at Anthropic, said on Tuesday that staff at the company are "genuinely frightened" about the trajectory of artificial intelligence. He warned that without a slowdown, the technology could pose an existential risk to humanity. The interview was conducted by the BBC from Coxon's London flat. His remarks add a personal voice to a growing chorus of insider concerns.
What Jacob Coxon Said and Who He Is
Jacob Coxon worked on safety‑focused projects at Anthropic, the San Francisco‑based AI startup known for its Claude language model. In a televised interview with the BBC on 10 September 2026, Coxon explained that a "strong chance" exists that AI could end humanity if development continues unchecked. He described a series of internal meetings where engineers expressed fear that the pace of model scaling was outstripping the team's ability to test for catastrophic failure modes. Coxon recalled a specific moment last month when a senior engineer asked, "Are we building something we can't control?" He said the sentiment was shared across teams in both the research and product divisions. The interview was recorded in his modest flat on Camden Road, where he displayed a whiteboard sketch of a neural network diagram to illustrate the speed of recent model upgrades. AI safety team members, according to Coxon, have begun documenting their concerns in a shared Google Doc that now contains over 200 entries. The BBC noted that Coxon left Anthropic in early 2025 after a disagreement over the company's risk‑assessment framework.
Why His Warning Matters to the Public
Coxon's comments matter because they come from someone who helped design the very systems that power chatbots, code generators, and recommendation engines used daily by millions. First, his warning highlights a gap between corporate optimism and internal risk assessments. While Anthropic markets Claude as a "helpful and harmless" assistant, Coxon says the engineering team worries about emergent behaviours that could be weaponised or cause large‑scale misinformation. Second, the concern is not limited to one firm. Similar unease has been reported at OpenAI, DeepMind, and other leading labs, suggesting a systemic issue in the AI industry. For ordinary people, this could translate into more frequent exposure to AI‑generated deepfakes, automated phishing attacks, or even algorithmic decisions that affect credit, employment, and legal outcomes without transparent oversight. Third, policymakers are watching these insider testimonies as they draft new regulations. The European Union's AI Act, slated for a vote in early 2027, may be reshaped by such revelations, potentially imposing stricter testing and documentation requirements. Job automation fears are also amplified when experts admit they cannot fully predict how quickly AI will replace certain roles, raising questions about future workforce stability. Coxon's remarks therefore serve as an early warning that could influence both consumer behaviour and legislative agendas.
“Jacob Coxon told the BBC that the staff at Anthropic are genuinely frightened because they see the gap between how quickly models are being released and how little we understand about their long‑term impacts.”
What Remains Uncertain About AI Risks
Despite Coxon's vivid description of internal fear, many technical details remain unknown to the public. Researchers have not disclosed the exact metrics they use to evaluate catastrophic failure, nor have they released the internal risk‑assessment models that guide deployment decisions. The precise timeline for when a potentially dangerous capability could emerge is also unclear; some experts argue it may be months, others claim years. Moreover, the extent to which external actors could exploit AI systems for malicious purposes has not been quantified. While Coxon mentioned a "strong chance" of existential risk, he did not provide a probability range, leaving the claim open to interpretation. The broader AI community continues to debate the alignment problem, the difficulty of ensuring that increasingly autonomous systems pursue human‑aligned goals. Until more transparent data are shared, policymakers and the public must navigate a landscape of speculation mixed with genuine insider concern.
Key Takeaways
- Jacob Coxon, former Anthropic researcher, warned that AI could pose an existential threat if development is not slowed.
- He said staff at Anthropic are genuinely frightened, documenting concerns in a shared internal document.
- Industry insiders across multiple AI labs share similar safety anxieties, suggesting a systemic risk perception.
- Policymakers in the EU and US are poised to act, with hearings and potential amendments to AI regulations imminent.
- Uncertainty remains about exact risk probabilities, timelines, and how quickly malicious actors could exploit AI.
What to Watch in the Coming Days
The next 24‑72 hours will likely see a flurry of reactions from both industry and regulators. Anthropic is expected to release a brief statement addressing Coxon's interview, which could include a clarification of its safety protocols or a denial of the alleged staff panic. At the same time, the U.S. Senate Committee on Commerce, Science, and Transportation has scheduled a hearing on AI safety for next Thursday, where former lab employees may be called to testify. In Europe, the European Parliament's AI oversight panel is set to publish a draft amendment to the AI Act that references insider warnings as part of its justification for tighter controls. Investors are also monitoring the story; Anthropic's parent company, a venture capital fund, may adjust its funding strategy if the perception of risk grows. Finally, social media platforms are likely to see a spike in discussions about AI safety, with hashtags such as #AISafety and #TechRisk trending. Observers should track official statements, legislative agendas, and market movements to gauge how Coxon's warning translates into concrete action. Regulatory hearings will be a key indicator of whether governments are moving from discussion to enforcement.
Anthropic allocated 15% of its 2024 R&D budget to safety research, according to an internal memo leaked to the Financial Times.
Jacob Coxon's testimony adds a personal dimension to the broader debate over artificial intelligence safety. While his fears are shared by many in the field, concrete evidence about imminent danger is still lacking. Governments, companies, and the public will need to balance innovation with precaution as the technology advances. Transparent dialogue and robust oversight may help bridge the gap between optimism and caution, ensuring that AI benefits are realized without compromising humanity's future.
