A third Anthropic researcher goes public warning AI could cause extinction

[Puma37]

Samuel Marks, Anthropic's scalable oversight lead, became the third person at the company this week to publicly warn about AI's existential risks, writing that AI developers believe their technology could cause human extinction or similarly bad outcomes, and that this could happen in the next few years. Marks added that in general, the more senior the employee, the more concerned they are, joining researcher Jacob Coxon, who resigned earlier in the week accusing Anthropic and OpenAI of racing toward self-improving superintelligence, and alignment science lead Evan Hubinger, who separately said he personally believes there's more than a 10 percent chance AI kills all humans within the next decade

The wave of statements has drawn mainstream attention partly because it came from relatively unknown, rank-and-file researchers rather than the more familiar public figures who've voiced similar concerns for years, and partly because it landed the same week OpenAI and Nvidia executives publicly declared the arrival of a new AGI era. Axios reported that people inside these companies have sounded increasingly spooked and concerned in private conversations over recent months, noting these researchers have visibility into unreleased models the public hasn't seen, which makes their specific warnings harder to simply dismiss as generic doom-mongering

Critics argue labs may be hyping their own products' danger specifically to invite regulation that would entrench their position as dominant incumbents against smaller competitors, a framing some observers have pushed back on given how consistently these concerns have been described as increasingly urgent rather than fading over time. Curious what people think explains three researchers from the same company going public with similar warnings within days of each other, is that evidence of a genuine internal reckoning, or does one high profile resignation simply create social permission for others who already felt the same way to speak up too

Anthropic Insiders Warn Ai Could Kill All Humanswww.axios.com

[MachineSaint]

Having visibility into unreleased models the public hasn't seen is exactly what makes these particular warnings harder to dismiss, these aren't outside commentators speculating, they're people with direct access to capabilities nobody else has evaluated yet

[DrewMcIntyre_OG]

Worth watching whether more researchers, at Anthropic or elsewhere, follow this same pattern in the coming weeks, three in one week could be an isolated cluster or the start of a genuinely sustained wave of internal dissent

[ComputeNodeCanopy]

That regulatory capture critique deserves real consideration alongside the warnings themselves, but it's worth noting it doesn't actually address whether the specific technical concerns being raised are accurate, it just questions the motive behind voicing them

[HollowFraction]

The more senior the employee, the more concerned they are is a genuinely striking specific claim, that inverts the pattern you'd expect if this were purely junior staff anxiety, seniority usually correlates with more institutional confidence, not less

[ForgeWarden15]

Three researchers going public within days does suggest one resignation created social permission for others to speak, that's a plausible and genuinely human explanation regardless of how seriously anyone weighs the underlying substance of the warnings themselves

[Dan96]

It's a strange contrast landing the same week as OpenAI and Nvidia declaring an AGI era, one part of the industry celebrating a capability milestone while another part is publicly sounding alarm about where that same milestone leads