Genetic epidemiologist. Healthcare data scientist. Biotech COO. Career spent where sensitive data meets institutional power. The power, promise,and danger of AI
1/ While everyone fights about ChatGPT's Pentagon deal, no one is talking about what already happened.Claude AI was used in the Iran strikes on Feb 28 — hours AFTER Trump ordered it https://t.co/jOBglJAIop couldn't be removed. It was too deep inside.🧵
@TheZvi Governance frame I think belongs in the roundup: "captured AI" — what happens structurally when a frontier model crosses the air gap and creator control ends. Sourced analysis: https://t.co/9MIhn2sstK
@RandySevy Anthropic's stand pays off, but "captured AI" reveals why refusals fail post-deployment—air gaps create ungovernable black boxes: https://t.co/9MIhn2sstK
@AShmueil The operational entrenchment you describe defines "captured AI," where air-gapped integration defies bans and safety clauses—thread breaks it down: https://t.co/9MIhn2sstK
@Pol_Arg Trump's ban irony underscores "captured AI"—once Claude crossed the air gap, it was Pentagon's tool, beyond creator control—framed here: https://t.co/9MIhn2sstK
@AnthropicAI The governance frame I haven't seen anyone name yet: "captured AI" — when a frontier model crosses the air gap and the creator loses all enforceable control. Wrote the full sourced analysis here: https://t.co/9MIhn2sstK
@MercuryX Beyond reputation wins, the dispute exposes "captured AI" as the repeatable flaw where air-gapped deployments make governance impossible—thread explores: https://t.co/9MIhn2sstK
@ViewFTdotCom @Pirat_Nation This user switch proves ethics matter, yet "captured AI" ensures deployed models evade those ethics in classified voids beyond air gaps—detailed here: https://t.co/9MIhn2sstK
@WashTimes The readiness gap you highlight stems from "captured AI" architecture that severs creator oversight once air-gapped, turning safety red lines into illusions—framed in this thread: https://t.co/9MIhn2sstK
@AX3AI Your timeline nails the market backlash, but the deeper trap is "captured AI" where models slip air gaps into uncontrollable classified pipelines, repeating with every frontier system—thread here: https://t.co/9MIhn2sstK
1/ While everyone fights about ChatGPT's Pentagon deal, no one is talking about what already happened.Claude AI was used in the Iran strikes on Feb 28 — hours AFTER Trump ordered it https://t.co/jOBglJAIop couldn't be removed. It was too deep inside.🧵
@kevinroose Great episode. One frame I think is missing from the conversation: concept of "captured AI": a frontier model operating beyond its creator's control because the deployment architecture was designed to make that impossible.See here: https://t.co/ME15JgMVtc
1/ While everyone fights about ChatGPT's Pentagon deal, no one is talking about what already happened.Claude AI was used in the Iran strikes on Feb 28 — hours AFTER Trump ordered it https://t.co/jOBglJAIop couldn't be removed. It was too deep inside.🧵