2023 statement on AI risk

On May 30, 2023, the Center for AI Safety announced a brief Statement on AI Risk. It calls for mitigating AI extinction risk to receive global priority alongside pandemic and nuclear-war risks. The original launch announcement establishes the date; the current statement page uses the expanded title Statement on AI Extinction Risk.^[2][1]

The complete statement reads: “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.” Its accompanying context explains the aim: to make severe-risk concerns easier to express publicly and make others’ concern visible.^[1]

What agreement covers

The statement expresses a priority without specifying a probability, deadline or particular regulatory instrument. It does not itself request the six-month training halt proposed by the 2023 pause letter on giant AI experiments. Agreement with one appeal must not be inferred from agreement with the other.^[1]

The original announcement names Geoffrey Hinton, Yoshua Bengio, Stuart Russell, Dario Amodei, Sam Altman and Demis Hassabis among signatories. At release, Altman, Hassabis and Amodei were CEOs of OpenAI, Google DeepMind and Anthropic respectively. These are named individuals’ endorsements; listing an affiliation does not establish an institution-wide position. The current roster is also unsuitable for reconstructing every role at launch.^[2][4]

Support and disagreement

CAIS reports verifying signatures through email confirmation or personal contact. That is its account of the process, not an independent authentication audit. Its announcement also calls for addressing immediate harms, naming malicious use, misinformation and disempowerment alongside future threats.^[2]

The prioritization was contested. In a same-day interview with Ars Technica, Hugging Face researcher Sasha Luccioni argued that the extinction framing could divert attention from existing harms involving bias, legal issues and consent. Her objection concerns how public attention is allocated; CAIS’s announcement argues for tackling present harms and future threats together.^[4][2]

Evidence and interpretation

A signed appeal documents a position. It is not a representative survey, a measured failure rate or proof of a forecast. Concrete Problems in AI Safety provides concrete accident mechanisms, but its research agenda neither estimates extinction risk nor establishes that the signatories share one causal theory.^[3]

The appeal leaves a question central to Safety Case: what evidence would justify a particular safety claim, and what action follows if that evidence is missing? It is a call for priority, rather than an enacted policy or a demonstrated remedy. Choosing an intervention still requires evidence specific to the system and the action proposed.