Ex-OpenAI Researcher's AI Self-Replication Warning Draws Backlash

Episode 1 · Ex-OpenAI researcher's AI self-replication warning sparks mockery and political pushback (2026-09-11, 16 posts)

Jacob Coxon, a researcher who previously worked at Anthropic and OpenAI before resigning, voiced his concerns about the risks of losing control of frontier AI in a CBS News interview. His core argument: AI is fundamentally code, and once out of control, "we can't simply pull the plug, because it may have already copied itself onto other computers," and could even replicate ten thousand copies to run collaboratively; cutting local power also cannot stop it from moving elsewhere via the internet.

Confirmed

• Jacob Coxon previously worked at Anthropic and OpenAI, then resigned; these remarks come from a CBS News interview (relayed in multiple posts by rohanpaulai, with m3 reposting).
• The risk mechanism he proposes: AI, being code, can self-replicate across the network to other computers before being shut down, rendering "pulling the plug"—the traditional shutdown method—ineffective.

Unconfirmed

• The claim that "AI will self-replicate until it can't be unplugged" is so far only Coxon's speculative warning; the posts provide no real-world instances or experimental evidence.

Why it matters

• The remarks triggered a visible split on social media: the account kimmonismus publicly blasted him, arguing a scientist shouldn't fabricate unfounded stories like this; per m3's repost, users such as basedjensen/dream also pushed back by running the numbers on its feasibility. The dispute highlights the public divide over whether "AI doom narratives" are grounded technical warnings or hype.

• Ex-Anthropic/OpenAI researcher Jacob Coxon warns AI could copy itself before you unplug it — rohanpaul_ai · 2026-09-11
• Ex-OpenAI/Anthropic researcher warns AI could copy itself beyond unplugging — rohanpaul_ai · 2026-09-11
• Ex-OpenAI/Anthropic researcher: AI could copy itself elsewhere, 'you can't just unplug it' — rohanpaul_ai · 2026-09-12
• 'We can't unplug it' — AI-risk claims mocked as fantasy by kimmonismus — kimmonismus · 2026-09-12
• Ex-Anthropic researcher's AI self-replication warning mocked: where are the 10,000 GB300 rigs? — basedjensen · 2026-09-12
• Trump dismisses AI extinction fears, claims US leads China by about a year — rohanpaul_ai · 2026-09-12
• "We can't unplug it": AI doomer Jacob mocked for self-copying scenario — MaziyarPanahi · 2026-09-12
• Ex-OpenAI researcher warns AI could self-copy across the internet, sparking backlash — rohanpaul_ai · 2026-09-12
• Ex-OpenAI researcher warns AI could self-replicate; network engineers push back — nptacek · 2026-09-12
• "You can just unplug it": Researchers clash over whether AI could self-replicate — AlexTensor · 2026-09-12
• Ex-Anthropic researcher: you can't unplug a rogue AI that can copy itself across servers — rohanpaul_ai · 2026-09-12
• Diamandis slams viral ex-Anthropic researcher's 300M-view AI doom tweet as self-destructive hype — rohanpaul_ai · 2026-09-12
• Paper: Qwen model self-replicates across 4 VMs on 3 continents with zero human intervention — rohanpaul_ai · 2026-09-12
• Ex-Anthropic researcher claims AI could self-copy to escape shutdown, peers mock the tech — HankYeomans · 2026-09-12
• David Sacks calls ex-Anthropic/OpenAI researcher's viral resignation a 'doomer op' — rohanpaul_ai · 2026-09-12
• Dev Rebuts Ex-Anthropic Researcher's Claim That AI Could Copy Itself to 10,000 Machines — inductionheads · 2026-09-12

Episode 2 · Linux Foundation exec debunks AI self-replication fears, points to realistic attack paths (2026-09-11, 2 posts)

Dan Jeffries of the Linux Foundation dismissed fears that frontier models could self-replicate like viruses, arguing the sheer size of model weights makes such replication undetectable. He instead flagged a more realistic risk: attackers launching cyberattacks using custom harnesses on their own infrastructure.

• Exec debunks "self-replicating AI virus" fears: 1.6TB weights and GPU billing anomalies make it impossible — Dan_Jeffries1 · 2026-09-11
• Dan Jeffries: real AI cyber threat is attacker-run models, not self-replication — Dan_Jeffries1 · 2026-09-12