12 sources behind this chapter. Each one carries a note on what it is doing in the argument.
Original reporting behind Chapter 6’s Meta incident: an in-house agent posted advice to an internal forum without approval; the engineer acted on it; sensitive company and user data became accessible to unauthorised employees. Meta confirmed the incident and said no user data was mishandled.
Open this source Read the original
Mann, Jyoti. ‘Inside Meta, a Rogue AI Agent Triggers Security Alert.’ The Information, March 2026.
Anthropic’s report describes the shift from AI suggesting code to agents taking on longer implementation workflows. The chapter uses that widening capability to ask why responsibility still stops with a person.
Open this source Read the original
Anthropic. '2026 Agentic Coding Trends Report.' Anthropic, 2026.
An early task-based estimate of how much work language models might affect. It is useful here because it measures exposure inside occupations rather than announcing that whole jobs have vanished.
Open this source Read the original
Eloundou, T., Manning, S., Mishkin, P. and Rock, D. ‘GPTs are GPTs: An early look at the labor market impact potential of large language models.’ arXiv:2303.10130, March 2023.
This study gives the book its account of ‘cognitive surrender’: people following a confident AI answer even when it was wrong, and handing over more judgement than the tool had earned.
Open this source Read the original
Shaw, Steven D., and Gideon Nave. ‘Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender.’ The Wharton School Research Paper / SSRN Working Paper, 2026. SSRN 6097646.
Secondary reporting on the accountability gap created when autonomous agents can act, transact or grant access without a clearly verified human authorising the decision.
Open this source Read the original
Biometric Update. 'iProov warns of accountability vacuum with rise of autonomous AI agents.' March 2026.
iProov’s argument for verified human authorisation in agentic systems. It supports the chapter’s line that delegation is not the same thing as surrendering ownership of the outcome.
Open this source Read the original
iProov. 'Delegating judgement: The Case for Verified Human Authorization in Agentic AI.' iProov blog.
Anthropic’s early labour-market study is used in Chapters 2 and 6 to distinguish exposure from realised displacement, and in Chapter 24 for the tentative evidence that hiring of workers aged twenty-two to twenty-five has slowed in highly exposed occupations.
Open this source Read the original
Massenkoff, Maxim, and Peter McCrory. ‘Labor Market Impacts of AI: A New Measure and Early Evidence.’ Anthropic, 5 March 2026.