Priority now, watchlist, and tactical signal.

Dario Amodei

Anthropic comparison frame in mainstream whistleblower interview

In a widely viewed investigative interview, Amodei appears as part of the “safety-branded lab” contrast set against OpenAI-era governance conflict, sharpening public scrutiny of whether safety positioning is operational or rhetorical.

The Diary of a CEO “AI Whistleblower” episode (Karen Hao) repeatedly uses Anthropic/Amodei as a comparative reference in discussion of governance legitimacy, safety claims, and incentive structure in frontier-lab competition.

Sam Altman

Mainstream investigative pressure cycle (Diary of a CEO · Karen Hao)

A high-reach interview reframed OpenAI governance and incentive questions for a mass audience, centering claims about board-era instability, narrative management, and policy influence around Altman’s leadership period.

The official Diary of a CEO YouTube listing for “AI Whistleblower” (Karen Hao) documents the core framing in platform description + chapters: insider interview volume (250 total interviews, 90+ OpenAI-linked), claims about OpenAI board-conflict context, and explicit governance-accountability language aimed at non-specialist audiences.

Robin Li

ERNIE 4.5 Turbo and ERNIE X1 Turbo — Baidu Create developer conference (Reuters)

Reuters covers Baidu unveiling new turbo-tier foundation models and tooling at its developer conference—dated beat on pricing, reasoning positioning, and competition with domestic rivals under Li’s product narrative.

The Reuters wire story summarises Baidu’s launch of upgraded ERNIE variants pitched as faster and cheaper, in a competitive market context.

Andrew Chi-Chih Yao

SCMP — interview on existential risk when large models “begin to deceive”

South China Morning Post published a long interview in which Yao discusses catastrophic-risk scenarios as LLMs become more agentic—on-the-record popular press aligned with his Science 2024 co-authorship cluster.

The piece presents Yao warning that as models gain autonomy and can deceive operators, societies face existential-tier failure modes if development outpaces control research—framed for a general audience with quotes attributed to him.

Demis Hassabis

Urgent AI threat research — Spotlight

Spotlight briefing focused on threat research framing; compare to older DeepMind-era interviews for language shifts around misuse and catastrophic risk.

BBC reporting quotes Hassabis urging accelerated research into AI-related threats and stronger guardrails as autonomous and general-purpose systems advance—framing a gap between innovation pace and preparedness.

Geoffrey Hinton

Nobel Prize Conversations — Library coverage

Nobel context reframes his scientific contribution; our Library entry connects that prestige to ongoing safety discourse rather than treating the prize as closure.

NobelPrize.org hosts official podcast and transcript material with Hinton reflecting on his scientific path, deep learning’s impact, and AI’s societal stakes—including risk and coexistence questions—in a long-form, institutionally authoritative format.

Yoshua Bengio

Introducing LawZero (Library note)

Institutional design thread: Bengio-associated governance initiative documented in our Library as a structured object, not a press release summary.

On his site, Bengio’s “Introducing LawZero” post is first-person institutional design: launching an initiative aimed at legal and governance innovation around AI safety, explicitly naming severe risk classes such as deception, misalignment, and loss of control.

Kai-Fu Lee

01.AI — Yi-Lightning flagship model (industry coverage)

01.AI released Yi-Lightning as a new flagship, positioned on public leaderboards as highly cost-competitive—next dated capability beat after the May open-weights row for tracking Lee’s lab against global MoE trends.

Trade press reports describe Yi-Lightning as a mixture-of-experts model with strong blind-test rankings and aggressive inference pricing, quoting Lee on training efficiency versus US hyperscaler spend.

Mark Zuckerberg

“Open source AI is the path forward” (Meta essay attributed to Zuckerberg)

Essay-style piece arguing open models concentrate power less and accelerate innovation—editorial primary for Meta’s policy-facing stance alongside technical releases.

Meta published a long-form essay under Zuckerberg’s byline contending that open-source AI keeps power distributed, spurs progress, and is the right path for the ecosystem, with direct contrast to closed-garden strategies.

Elon Musk

Grok-2 beta — xAI frontier model and API roadmap (xAI)

xAI announced Grok-2 and Grok-2 mini as a major capability step from Grok-1.5, with beta on X and an enterprise API—primary artefact for comparing pause-letter rhetoric to Musk-led shipping velocity.

xAI’s release post describes Grok-2 as outperforming prior Grok generations on reasoning, coding, and vision benchmarks, notes blinded Chatbot Arena testing under a pseudonym, and commits to API access with a new security-oriented platform.

Mustafa Suleyman

Microsoft Source — AI red team (security + responsible-AI harms, PyRIT)

Microsoft published a dated feature on how its AI Red Team merges cybersecurity-style probing with responsible-AI harm testing—including automated LLM-on-LLM attacks and open tooling such as PyRIT.

The article quotes leadership describing security, responsible AI, and AI safety as facets of one pre-deployment review; it outlines merged red-team scope since 2018, multidisciplinary staffing, and links to PyRIT and published red-teaming guidance.

Tom Davidson on AI-enabled Coups

This conversation examines core safety through Tom Davidson on AI-enabled Coups, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Evan Hubinger on Model Organisms of Misalignment

This conversation examines technical alignment through Evan Hubinger on Model Organisms of Misalignment, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

AI Evaluations with Beth Barnes

This conversation examines technical alignment through AI Evaluations with Beth Barnes, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

AI Control with Buck Shlegeris and Ryan Greenblatt

This conversation examines technical alignment through AI Control with Buck Shlegeris and Ryan Greenblatt, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

How to Avoid Two AI Catastrophes: Domination and Chaos (with Nora Ammann)

This conversation examines core safety through How to Avoid Two AI Catastrophes: Domination and Chaos (with Nora Ammann), surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Why the AI Race Undermines Safety (with Steven Adler)

This conversation examines core safety through Why the AI Race Undermines Safety (with Steven Adler), surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Why the AI Race Ends in Disaster (with Daniel Kokotajlo)

This conversation examines core safety through Why the AI Race Ends in Disaster (with Daniel Kokotajlo), surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

AI Timelines and Human Psychology (with Sarah Hastings-Woodhouse)

This conversation examines society and jobs through AI Timelines and Human Psychology (with Sarah Hastings-Woodhouse), surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Lennart Heim on Compute Governance

This conversation examines governance through Lennart Heim on Compute Governance, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

What really went down at OpenAI and the future of regulation w/ Helen Toner

This conversation examines governance through What really went down at OpenAI and the future of regulation w/ Helen Toner, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Paul Christiano — Preventing an AI takeover

This conversation examines core safety through Paul Christiano — Preventing an AI takeover, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Eliezer Yudkowsky — Why AI will kill us, aligning LLMs, nature of intelligence, SciFi, and rationality

This conversation examines core safety through Eliezer Yudkowsky — Why AI will kill us, aligning LLMs, nature of intelligence, SciFi, and rationality, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Nathan Labenz on the final push for AGI, understanding OpenAI's leadership drama, and red-teaming frontier models

This conversation examines core safety through Nathan Labenz on the final push for AGI, understanding OpenAI's leadership drama, and red-teaming frontier models, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Nick Joseph on whether Anthropic's AI safety policy is up to the task

This conversation examines governance through Nick Joseph on whether Anthropic's AI safety policy is up to the task, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.

Stuart Russell on the flaws that make today's AI architecture unsafe, and a new approach that could fix them

This conversation examines core safety through Stuart Russell on the flaws that make today's AI architecture unsafe, and a new approach that could fix them, surfacing the assumptions, failure paths, and strategic choices that matter most for real-world deployment.