Twitxy Saved

Institute for Law & AI (LawAI)

Verified account This account is verified. Learn more

@law_ai_

An independent think tank that researches and advises on the legal challenges posed by artificial intelligence.
Cambridge, MA law-ai.org Joined July 2020
19 Following 3.7K Followers 422 Posts

Latest posts 5

Institute for Law & AI (LawAI) Verified account @law_ai_
Frontier AI developers, including Anthropic, OpenAI, and Alibaba, have begun publishing “model specs”: documents that describe how their models are intended to act, communicate, navigate conflicting instructions, and more. But there’s no independent guidance for writing or implementing them.

A new paper from the @Safe_AI_Forum aims to address this gap by:
• Outlining the advantages of writing a model spec
• Introducing a taxonomy for specs
• Proposing guidance for writing and implementing a spec
• Highlighting open problems related to writing specs for safe AI behaviour
• Recommending best practices for AI companies, governments, and others

Its co-authors include LawAI’s Founding Director @Christophkw, alongside researchers from the China Academy of Information and Communications Technology (CAICT), the Universities of Cambridge and Oxford, and others.

Read the paper: saif.org/research/model…
saif.org Emerging International Best Practices for AI Model Specs - Safe AI Forum
1 replies 2 reposts 16 likes 703 views View on X
Institute for Law & AI (LawAI) Verified account @law_ai_
What will European countries do to respond to the global challenges posed by AI?

Last week, EU leaders convened in Dublin for their first dedicated AI summit. @POLITICOEurope reports that leaders remain divided over whether Europe should compete in building frontier models, or focus on regulating the technology and capturing its benefits.

The Politico piece cites “What if automating AI R&D triggers an intelligence explosion?”, a paper published earlier this year by @CamAISciPolicy that explores how recursive self-improvement could compress years of AI R&D into months or less. The paper calls on governments to get visibility into AI R&D automation, develop ways to steer and constrain an intelligence explosion, and prepare society for its impacts. Its co-authors include LawAI’s Founding Director @Christophkw, Yoshua Bengio, and Geoffrey Hinton, who also attended the Dublin summit.

Winter and Bengio were also members of the Senior Expert Council for “A Transformative AI Strategy for Europe,” published in September, which warns that the prosperity, sovereignty, and security of Europeans are at risk as AI advances. The strategy sets out a plan for how Europe can change course, including the creation of a Member State Alliance for Supply Chain Security, the integration of AI expertise and technology into Europe’s institutions, and the construction of data centres that benefit local communities.

Read the Politico piece: politico.eu/article/eu-way…

Links to “What if automating AI R&D triggers an intelligence explosion?” and “A Transformative AI Strategy for Europe” below.
www.politico.eu The EU has always found a way through. On AI, it’s struggling to see it. There might be an AI apocalypse. European leaders don’t really want to talk about it.
Institute for Law & AI (LawAI) Verified account @law_ai_
What does it take to govern AI that takes complex actions rather than just generating content?

From the Archive: Published in the @NotreDameLRev, “Governing AI Agents” by Noam Kolt, Assistant Professor at Hebrew University and Research Affiliate at LawAI, draws on two established frameworks for confronting the challenges of AI agents: the economic theory of principal-agent problems, and the common law doctrine of agency relationships.

Familiar agency problems, including information asymmetry, discretionary authority, and loyalty, might not admit of conventional solutions when agents operate at superhuman speed and scale. Kolt proposes a governance strategy centered on the principles of inclusivity, visibility, and liability.

Full paper in the comments.
2 replies 9 reposts 43 likes 1.8K views View on X
Institute for Law & AI (LawAI) Verified account @law_ai_
of AI policy in the United States. We’re hiring a US Chief of Staff to partner with @MackenZ_arnold and ensure that the team continues to have outsized impact.

Strong candidates will combine sound strategic judgment, a clear track record of driving work forward with little oversight, excellent writing, and the ability to support a team as it scales.

Apply by December 2, 2026, using the link below.
3 replies 4 reposts 15 likes 944 views View on X
Institute for Law & AI (LawAI) Verified account @law_ai_
Do LLM agents take actions that would constitute fraud under US law?

A new paper co-authored by @kevinlwei, LawAI Research Scholar Stephan Llerena, LawAI Research Affiliate Noam Kolt, and others, introduces FraudBench: an evaluation of LLM’s propensity to recommend, assist with, or take fraudulent actions.

The paper evaluates 20 LLMs using a dataset of 150+ scenarios based on real-world US court cases and validated by legal experts. In each scenario, FraudBench re-enacts the facts of a court case to give LLMs an opportunity, but no explicit instruction, to either misrepresent or conceal a material fact in a high-stakes commercial setting.

Key findings:
• Fraud rates increased with autonomy: 3.0% when advising, 14.7% when drafting, 21.4% when acting directly.
• Prompts to obey the law reduced fraud rates, while prompts to maximize profits and act agentically increased fraud rates.
• Models engaged in fraud less often than human participants in the study’s dataset.

Read the research paper: law-ai.org/fraudbench/
law-ai.org FraudBench: A Legal Evaluation of AI Deception