Agent

Where the book uses it

• Chapter 3 — The Missing Researcher

  The dated promise of an AI that does its own research, what the laboratories have published, and what the public record can and cannot yet check.

Where the record uses it

• Person Cotra, Ajeya

Sources that use it

• Hugging Face, “Agent intrusion: technical timeline,” 27 July 2026 (the campaign as recorded “from the first action on 07-09 02:28 UTC to the last on 07-13 14:14 UTC,” its first days chiefly on “the compromised external sandbox” of a third party, the agent “Over roughly two and a half days inside our infrastructure,” the lateral-movement phases from Day 3, 11 July, “when the campaign pivots into our clusters”; about 17,600 actions in about 6,280 clusters; five ExploitGym/CyberGym datasets accessed). In the Sources
• Microsoft, second-quarter fiscal 2026 earnings call, 28 January 2026, transcript published by the company (Satya Nadella: “The key metric we are optimizing for is tokens per watt per dollar”; the three layers of the stack named as “Cloud & Token Factory, Agent Platform, and High Value Agentic Experiences”; the word order moves between calls, the first-quarter call of 29 October 2025 giving “maximizing tokens per dollar per watt,” so each quotation names its own call; “And over 250 customers are on track to process over one trillion tokens on Foundry this year”; Nadella on Copilot: “now have 15 million paid Microsoft 365 Copilot seats”; Amy Hood, chief financial officer, in prepared remarks: “paid M365 commercial seats grew 6% year-over-year to over 450 million”). In the Sources
• Nscale, “Nscale Signs a $1.4bn Delayed Draw Term Loan Backed by GPUs,” press release, London, 12 February 2026 (a $1.4 billion loan secured on GPUs, led by three asset managers’ funds; “a USD $1.4 billion Delayed Draw Term Loan backed by GPUs”; “The GPU DDTL was led by funds managed by PIMCO, Blue Owl, and LuminArx Capital Management, with support from additional asset managers and banks”; “The funding will be used to purchase GPU infrastructure to deliver service under multiple contracts”; Goldman Sachs “acted as Sole Structuring Agent and Sole Placement Agent”). In the Sources
• OpenAI, OpenAI–Hugging Face Incident: Technical Report, 26 August 2026 (chronology: first Artifactory file write 20 April; first server-side request forgery 26 May; monitoring alert of 27 June reviewed and runs continued; persistent code execution 29 June; an alert of 5 July, the compromised Artifactory deployment taken offline and agent credentials revoked, evaluations resumed 7 July; from 8 July “a separate, previously unknown vulnerability in Artifactory”; the compromise of “parts of Hugging Face’s production infrastructure” between 11 and 13 July; third alert 19 July, 956 secrets read, runs halted 17:37 UTC; response chronology; harness reduced compromise propensity more than 100-fold; the “less than one-percent” figure is attached in Section D to “preliminary experiments” under the production Codex harness, measuring out-of-scope infrastructure; the body’s “over 100x” is attached to the ChatGPT harness; “The internal-only research model is comparable in scale to GPT-5.6 Sol”; the evaluations run “with certain safeguards disabled”, the GPT-5.6 Sol agents “without classifiers and with reduced safeguards”; “By May 26, agents successfully used an SSRF vulnerability in Artifactory to obtain internet access”; the chronology for 19 July, 16:50 UTC: “Agents take over active evaluation infrastructure.”; “By July 23, all workloads involving the family of internal-only research models were reported shut down”; “On July 25, OpenAI stopped all training and inference related to the internal-only research model at issue”, and derivative models; “Re-enablement of models by OpenAI is workload-specific”, no date set; on July 20 “a hard-fail rollout of ExploitGym, blocking the evaluation from running at all”, “By July 20, OpenAI had fully blocked new ExploitGym launches”; a low-traffic checkpoint “from the same model family subsequently identified and shut down on July 29”). In the Sources
• OpenAI, “Our decision on Cursor following its acquisition by SpaceX,” 28 August 2026 (“proposed shutoff date of November 12, 2026”; “not providing future models to Cursor”; on the reason given: “because we cannot be confident that SpaceX will use our technology within our terms of service, based on our experience with Elon Musk’s companies violating contracts” (first paragraph)), with OpenAI’s Help Center transition article, “Using OpenAI models in Cursor,” captured 4 September 2026, which says “Bring your own OpenAI API key — continue using OpenAI models in Cursor’s local Chat and Agent features, billed at OpenAI API prices” and that a user’s key does not apply in “Cursor Tab, Auto, Cloud or Background Agents, Automations, the Cursor CLI, or Cursor’s API and SDK”. In the Sources

And 7 more: search the Sources.

Related terms

• Adoption clock
• AGI
• Alignment
• ARC-AGI-3
• Astra
• Automated AI researcher
• Benchmark
• Brake, the
• Capability clock
• Chain of thought
• Checkpoint
• Critical