Briefings
9 results
Anthropic Questions
This investor briefing analyses what’s new in Anthropic’s S-1 for the nine topics covered in our recent Investor Guide to Frontier AI Governance. It includes questions for investors to ask Anthropic across: AI R&D automation, incidents, corporate governance, third-party evaluation, and access to compute.
Investor Guide to Frontier AI Governance
The risks from developing advanced AI, and the safety practices to mitigate these risks, are highly material to investors in frontier AI developers. However, these topics are undercovered by mainstream investment research. This report addresses this gap by outlining the investment impacts from frontier AI risks (section II), resources investors can draw on to assess and engage frontier AI developers (section III), and the key topics and questions to consider (section IV).
AI Scenarios
AI scenarios are a key tool for investors to use in company analysis, investment decisions, stewardship, and portfolio construction. This briefing outlines the current resources investors can draw on and key questions to consider.
AI forecasts
Summary notes from an investor expert discussion with Josh Rosenberg the CEO of the Forecasting Research Institute on AI forecasts. A key challenge for investors on AI governance is the incredibly varied forecasts on AI capabilities and impacts. This may be one reason why many investors are struggling to consider the more uncertain and disruptive risks from AI, and why stewardship teams are instead focusing on how it impacts existing sustainability topics. To help address this we held a discussion with Josh Rosenberg from the Forecasting Research Institute on their research analysing AI predictions from 250+ subject matter experts and highly accurate forecasters. Three points stood out: • Heads of labs expect the greatest and fastest impacts, next are experts and superforecasters, and lagging are the expectations of the public. • The average expert probability of AI being equivalent to a technology of the century (e.g. electricity) is 60%, and 30% for being a technology of the millennium (e.g. agriculture). • Predictions on workforce impact diverge sharply between economists, AI experts, and the public.
AI safety state of play
Summary notes from an investor expert discussion with Stephen Clare on the International AI Safety Report 2026. Stephen Clare was one of the two lead writers for the recent report – International AI Safety Report 2026. The report covers 1) current model capabilities and likely progress by 2030, 2) AI risks from malicious use, malfunctions, and systemic impacts, 3) institutional and technical risk management. This is the most credible review of these topics and is closest equivalent to the IPCC for climate. Key points from the discussion: • Over the last year AI risks have moved from theoretically possible to real-world examples. • Self-governance and voluntary commitments have been surprisingly successful, but the competitive pressure to launch models quickly is intense and limits the focus on safety. More external pressure would help, including from investors. • A key challenge for AI governance is the declining effectiveness of model evaluations – the evaluation gap. Models increasingly recognise they are being tested and adjust their behaviour. • Investors can help by pushing developers for longer pre-deployment testing windows and deployers to only purchase AI systems that have been externally evaluated.
AI audits and verification
Summary notes from an investor expert discussion with Miles Brundage from AI Verification and Evaluation Research Institute on third-party audits of AI systems. AI audits are a critical component of AI governance, just as independent verification is a key component in many industries. Investors can play an important role in encouraging companies to conduct third-party audits and assessing whether they engage with external evaluation. This can help investors improve their investment decisions by understanding which companies have effective AI risk management. The challenge is the AI audit ecosystem is nascent with more work needed on standards and more providers, and there is not yet an easy way for investors to see which companies are conducting AI audits. Even so, investors can ask questions and support the growth of this ecosystem. AVERI focuses on frontier AI audits, but there is also an important role for AI audits for deployers. Investors can ask deployers if they have conducted an AI audit and if they only purchase AI systems that have had an external evaluation.
AI legal liability
Summary notes from an investor expert discussion with Gabriel Weil from the Institute for Law & AI and the Touro University Law Center on AI legal liability. AI legal liability is central to investment materiality but remains uncertain and evolving with court cases and regulation. Three points from the discussion: • Third-party harms are likely to create larger liabilities than harms to customers. Taking reasonable precautions is critical for AI developers to limit their potential liability from releasing a model that could be used for harm. • Regulations are progressing, but fragmenting. The state-by-state picture is splitting between liability shields (Illinois SB 3444) and liability expansion (NY A8833, RI HB 5224). California SB 53 and NY's RAISE Act are now law and non-compliance can be treated as evidence of negligence in subsequent injury cases. • Investors can start by asking companies if they have assessed their AI legal liability and how they are managing this with third-party evaluation and insurance.
Multi-agent risks
Summary notes from an investor expert discussion with Marta Bienkiewicz from Cooperative AI on multi-agent risks from advanced AI. Key takeaways for investors: • Encourage companies to be cautious deploying multi-agent systems as we cannot fully control and oversee these systems yet – the root cause of failures cannot be identified by human annotators in up to 30% of simulated scenarios and model drift means a compliant system today may not be tomorrow. And regulatory compliance and legal liability are unclear. • A key part of the solution is developing an ecosystem of AI auditors to provide external assurance. Investors can support this by encouraging companies to commission third-party audits rather than relying on developers' self-certification. • Encourage AI developers to build cooperative intelligence into agent design, train agents to understand legal and social norms, and develop circuit-breaker capabilities to halt network-level failures.
AI risk management
Summary notes from an investor expert discussion with Malcolm Murray the research lead at SaferAI on AI risk management. Malcolm leads research at SaferAI which is a non-profit producing research, tools, and assessments on AI risk management. SaferAI produces one of the only assessments of AI model developers on risk management – SaferAI Ratings. The session highlighted that AI risk management is falling behind model capabilities, but investors can help by pushing companies to adopt existing best practices. • The gap between predicted AI risks and current real-world harm is still wide, but it will probably close soon. Cyber could be first for a major AI incident (as highlighted by Mythos recently) and labour market disruptions are steadily increasing. • Current AI risk management practices are lacking in key areas such as novel risk identification (for example, manipulation and multi-agent risks) and board oversight. Frontier safety frameworks have a high degree of discretion and approaches are weakening over time. • On the positive side, if companies adopted best practices across SaferAI’s 65 sub-criteria there would be a significant improvement from a median score of 18% to 51%. Hence, investors can play an important role in encouraging companies to adopt these.