发生了什么
The BBC reports that Jacob Coxon, a former Anthropic and OpenAI researcher, warned that AI researchers fear the pace of development and its possible effects on humanity. He supported calls for slower development, regulation and independent monitoring, while Anthropic defended its safety work. The report adds new reactions to Anthropic CEO Dario Amodey’s existing call for greater caution.
The BBC reports that Coxon, 27, said AI researchers are “genuinely afraid” of the current pace of progress and argued that humanity could face a very high risk of dying in the near future if development is not slowed. He welcomed Amodey’s proposal for industry-wide restraint but said China would need to be involved to avoid an international race. These are Coxon’s views as reported by the BBC, not independently established forecasts.
The report connects Coxon’s comments to an essay by Anthropic CEO Dario Amodey, published on the 12th according to the source, which argued that AI development itself is not the problem but that companies and governments need time to address serious risks. The BBC also reports support for slowing development, regulation or independent monitoring from OpenAI CEO Sam Altman and xAI CEO Elon Musk.
Anthropic told BBC News that it has consistently acknowledged both major benefits and unprecedented risks from AI. Its spokesperson said the company is developing strong safeguards, researching how models operate, testing models for misalignment risks and publishing results. The source provides no independent assessment of those safeguards.
The BBC reports additional warnings from Anthropic scientist Evan Hubinger and computer scientist Geoffrey Hinton, who each gave personal estimates about the possibility of human extinction. It also reports criticism from Hugging Face CEO Clément Delangue and Faculty CEO Marc Warner, who said such probabilities are difficult to establish. The source contains no technical evidence validating any of these estimates.
为什么这很重要
The report matters because it documents a widening public disagreement inside and around leading AI companies over whether development is moving faster than safety work and government oversight. The practical proposals described are slower deployment of powerful models, independent monitoring and international coordination, including with China. However, the most severe claims are forecasts or personal estimates, not independently demonstrated findings. The source does not establish that extinction is imminent, quantify a verified risk, or show that any government or company has adopted new restrictions.
This is a meaningful AI-safety development because a former employee’s public account is presented alongside a company leader’s call for slower development and safeguards. Together, the statements show that concerns are not limited to outside critics, although the report does not measure how representative these views are among AI researchers.
The practical implication is that safety arguments could increasingly affect release governance, regulation and international coordination. Those consequences remain prospective: the source does not report a new law, agreement, deployment pause or independently verified incident. No product access conditions or pricing are relevant or documented.
互动机制：它实际上是如何运作的
以交互方式探索这一发展背后的基础技术。
Why can ethical evaluation not be reduced to one model score?
接下来看什么
Watch for concrete policy or company actions following the calls for restraint: formal monitoring requirements, deployment limits, international agreements, or changes to model-release processes. The source does not document a new product, access change, price, safety test result or binding policy. It also does not independently verify the extinction probabilities cited by interviewees or establish the timing of the reported interviews relative to the current 96-hour window.
Whether Anthropic or other AI companies turn calls for restraint into verifiable practices, such as independent model monitoring, published evaluation thresholds or slower releases.
Whether governments respond with specific rules or international discussions rather than general statements about AI competition and risk.
Whether independent technical research supports, narrows or challenges the severe risk estimates quoted in the report. The source does not provide evidence sufficient to resolve that dispute.
The source text does not include a publication date, so the recency of the underlying interviews cannot be independently confirmed from the supplied material.