AI Policy Bulletin

AI Policy Bulletin publishes peer-reviewed analysis on AI governance – made accessible for busy policymakers. Got an AI policy idea? Pitch us.

[AI Policy Bulletin @AIPB_org]

Read it here: https://t.co/KL5P00WpQb

[AI Policy Bulletin @AIPB_org]

New AIPB article: Europe Should Bet on Narrow AI

Instead of trying to catch up in general-purpose AI, Europe should play to its industrial strengths, writes Tina Wünn @pourdemain_ngo.

[AI Policy Bulletin @AIPB_org]

@MITAIRisk Read more here: https://t.co/PGZ30iT6F0

[AI Policy Bulletin @AIPB_org]

New research from @MITAIRisk surveyed 272 AI experts about AI risks.

It found experts believe 18 of 24 risk domains carry more than a 10% chance of catastrophe within 5 years, under business as usual.

Catastrophe means things like over 1 million deaths or $100B in losses.

[AI Policy Bulletin @AIPB_org]

.@MITAIRisk suggest three implications:

1. The size of the risks means that substantial mitigations are needed.

2. Relying on AI developers' voluntary action alone is insufficient – AI needs rules and enforcement.

3. Some risks may need international coordination.

[AI Policy Bulletin @AIPB_org]

Does your org work on AI evals?

@NIST is taking letters of interest to collaborate on AI standards.

Its AI Consortium will have 6 task groups, including:
- AI Testing, Evaluation, Verification & Validation
- Annotation for AI Risks & Validity
- Chemical & Biological Security

[National Institute of Standards and Technology @NIST]

NIST is extending the scope of and calling for new members for the newly renamed NIST AI Consortium. It will focus on AI innovation and adoption, with six task groups concentrating on aspects of AI measurement science and evaluation.

Details at https://t.co/t35K755zn7

[AI Policy Bulletin @AIPB_org]

Read it here: https://t.co/INXCJgNXzZ

[AI Policy Bulletin @AIPB_org]

New AIPB article: AI Cybersecurity Needs an Anti-Money Laundering Playbook

Disconnected prompts can add up to sophisticated cyberattacks. US regulators need a way to see the full picture, writes Jamie Johnson @ERA_Cambridge.

[AI Policy Bulletin @AIPB_org]

Even without transformative AI, much of the world's compute buildout through the 2030s could happen in space – as per new research by @finmoorhouse & Avi Parrack.

Worth watching, particularly for policymakers hoping to shape AI through leverage over domestic data centers.

[Forethought @forethought_org]

New post: Will we really put data centers in space?

Read it here: https://t.co/b03A1fZBjf

[AI Policy Bulletin @AIPB_org]

Another opportunity to input into EU AI Act guidelines – this time on high-risk AI systems.

If you work on AI in areas like hiring, biometrics, healthcare, or law enforcement, you have until 23 June to tell the @EU_Commission what you think.

[AI Policy Bulletin @AIPB_org]

Any AI policy researchers working on transparency and deepfakes?

The @EU_Commission has an open consultation on AI transparency guidelines, closing 3 June.

[AI Policy Bulletin @AIPB_org]

A new report today from the UK government's @AISecurityInst judges it 'almost certain' that AI systems will increasingly recognise when they're being tested and adjust their behaviour.

Makes you wonder - what will this mean for policies built on pre-deployment testing? 🤔

[AI Security Institute @AISecurityInst]

The safety of advanced AI systems increasingly depends on the ability to oversee them. Our new report examines today’s AI oversight landscape, finding many pathways likely to lead to its degradation.🧵

[AI Policy Bulletin @AIPB_org]

👀

"The draft, in its current form, calls for a "voluntary framework" to be established under which AI labs would share their models with the government at least 90 days before public release and also give access to certain critical infrastructure providers."

[Ashley Gold @ashleyrgold]

Scoop: Trump AI executive order seeks early government access to frontier models https://t.co/IeOnPOnWC9

[AI Policy Bulletin @AIPB_org]

https://t.co/iaJlN2c6ZW

[AI Policy Bulletin @AIPB_org]

"There is a world of difference between building a model and serving it to the planet."

@RyanFedasiuk @AEI argues that China's compute shortages show US export controls are largely working - now the US needs to plug the gaps.

[AI Policy Bulletin @AIPB_org]

https://t.co/x8zT1xvIWr

[AI Policy Bulletin @AIPB_org]

Any AI policy researchers working on transparency and deepfakes?

The @EU_Commission has an open consultation on AI transparency guidelines, closing 3 June.

[AI Policy Bulletin @AIPB_org]

Read the article: https://t.co/dloS6dpuhq

[AI Policy Bulletin @AIPB_org]

"When much of today's AI is not owned but permissioned, an AI strategic reserve might be an answer. Most middle powers lack one."

In their AI Policy Bulletin article, @JakiKasia and @Charles_Mrt explain why governments need a plan for when access to frontier AI is cut off.

[AI Policy Bulletin @AIPB_org]

@CommerceGov https://t.co/1DeVHzmIQ4

[AI Policy Bulletin @AIPB_org]

The US is taking action against distillation attacks – a technique Chinese AI companies have been documented using to catch up to their American competitors.

In this @iapsAI policy memo, @theobearman looks at what the US is doing so far and what's still missing.

[AI Policy Bulletin @AIPB_org]

His key recommendations:
• @CommerceGov should consider distillation risks when making decisions about chip exports & remote chip access.
• Amend the Deterring American AI Model Theft Act to fix gaps on industry cooperation, info sharing, trade secret theft and more.