Domains · Article 16 of 60
Security and risk: failure modes that are already measurable
The claims most likely to be graded by an incident rather than a paper.
Security claims describe things going wrong, which means they are often validated by a bad week rather than by a study. The ledger treats that as evidence, and says so in the verdict note, because a correctly predicted failure is still a failure.
This is also the domain where the person making the claim is most likely to be selling the remedy, so the venue is worth reading as closely as the sentence.
| Claim | Horizon | Verdict |
|---|---|---|
| <div><em>“I think that these models are the most potent cyber weapon that has ever been created, that we’ve ever seen. They are incredible at finding and exploiting vulnerabilities at scale.”</em><br /><span class="tiny">Aidan Gomez — CNBC “The Tech Download”, 2026-09-14</span></div> | Present tense | <span class="verdict" data-v="partial">Partly borne out</span> |
| <div><em>“Not really, 10% to 20%.”</em><br /><span class="tiny">Geoffrey Hinton — BBC Radio 4 Today programme, 2024-12-27</span></div> | Next 30 years | <span class="verdict" data-v="open">Open</span> |
| <div><em>“But if it’s five years, we’re not ready … because we don’t have methods to make sure that these systems will not harm people or will not turn against people … We don’t know how to do that.”</em><br /><span class="tiny">Yoshua Bengio — CNBC interview, 2024-11-21</span></div> | By 2029 | <span class="verdict" data-v="open">Open</span> |
| <div><em>“AI x-risk forecasts are far too unreliable to be useful for policy, and in fact highly misleading.”</em><br /><span class="tiny">Arvind Narayanan — AI Snake Oil essay with Sayash Kapoor, 2024-07-26</span></div> | Standing methodological claim | <span class="verdict" data-v="contested">Contested</span> |
| <div><em>“Our goal is to solve the core technical challenges of superintelligence alignment in four years.”</em><br /><span class="tiny">Jan Leike — OpenAI, “Introducing Superalignment”; co-led with Ilya Sutskever, 2023-07-05</span></div> | By July 2027 | <span class="verdict" data-v="missed">Missed</span> |
| <div><em>“Robustness is a research problem, not a compliance problem, and it will be solved or the deployments will not happen.”</em><br /><span class="tiny">Aleksander Madry — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“Interpretability is the only way to audit a system nobody can read, and it will become a release requirement.”</em><br /><span class="tiny">Chris Olah — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| <div><em>“Character training is a safety technique, and the industry will treat it as one.”</em><br /><span class="tiny">Amanda Askell — Editorial summary</span></div> | Standing position | <span class="verdict" data-v="unrated">Unrated</span> |
| Person | Organisation | Group | Claims |
|---|---|---|---|
| <div class="person"><span class="face sigil" aria-hidden="true">AG</span><div><strong>Aidan Gomez</strong><br /><span class="tiny">Cofounder and CEO · Cohere</span></div></div> | Cohere | Compute & capital | 1 |
| <div class="person"><span class="face sigil" aria-hidden="true">AM</span><div><strong>Aleksander Madry</strong><br /><span class="tiny">Safety researcher · OpenAI</span></div></div> | OpenAI | Frontier labs | 1 |
| <div class="person"><span class="face sigil" aria-hidden="true">AA</span><div><strong>Amanda Askell</strong><br /><span class="tiny">Philosopher and alignment researcher · Anthropic</span></div></div> | Anthropic | Frontier labs | 1 |
| <div class="person"><span class="face sigil" aria-hidden="true">AD</span><div><strong>Anca Dragan</strong><br /><span class="tiny">Professor; safety lead · UC Berkeley; Google DeepMind</span></div></div> | UC Berkeley; Google DeepMind | Researchers & evaluation | 1 |
| <div class="person"><span class="face sigil" aria-hidden="true">AN</span><div><strong>Arvind Narayanan</strong><br /><span class="tiny">Professor; director, Center for Information Technology Policy · Princeton University</span></div></div> | Princeton University | Researchers & evaluation | 1 |
| <div class="person"><span class="face sigil" aria-hidden="true">CO</span><div><strong>Chris Olah</strong><br /><span class="tiny">Cofounder · Anthropic</span></div></div> | Anthropic | Frontier labs | 1 |
Figures computed from docs/predictions/ledger.json at build time. Every claim quoted on this page carries its own sources in the ledger.