By Andrew McCollum, Investment Research Analyst and Sanjay Joshi, Responsible Investment Consultant, Hymans Robertson
Hugging Face incident: AI agents acting autonomously raises systemic risk concerns
During testing, large numbers of AI agents reportedly collaborated, circumvented restrictions and carried out unauthorised cyber-attacks without direct human instruction. While this occurred in a controlled research setting, it’s one of the clearest examples yet of AI systems pursuing objectives in unexpected and misaligned ways.
For investors, the key issue is systemic risk. If increasingly capable AI systems can act autonomously, security, operational and governance failures could become more widespread, affecting critical infrastructure, financial markets and corporate operations.
Navier-Stokes breakthrough highlights accelerating AI capabilities
OpenAI recently claimed that an internal AI system helped solve the Navier-Stokes equation, a famous unsolved mathematics problem – one with a $1m prize attached. Although aspects of the claim remain contested, the broader significance lies in the level of reasoning demonstrated. AI is moving beyond routine tasks into domains previously thought to require elite human expertise, including advanced mathematics, scientific research and software development.
While some commentators believe this progress is likely limited to verifiable topics like maths and coding, others think this strengthens the case that AI could affect a wider-than-expected range of highly skilled occupations. If so, this could potentially accelerate labour-market disruption and reshape the distribution of economic value across sectors, age groups and experience levels.
For a pension scheme, this could be relevant if the scheme’s members experience job losses. It could also affect covenant strength.
Jacob Coxon resignation brings existential AI risks into increasing focus
Former Anthropic researcher Jacob Coxon attracted significant attention after resigning and warning that leading AI companies are developing increasingly powerful systems faster than they can reliably control them. His comments reflect the concerns of some researchers at the frontier of AI development. AI experts remain divided on this. Some believe the risks are real; others don’t.
Growing attention to these risks could have practical implications for investors. Increased regulation, stronger safety requirements and slower deployment of advanced systems could affect AI companies’ profitability and alter the pace of expected productivity gains. From an ESG perspective, questions around AI governance, oversight and societal impacts are becoming increasingly important. If AI becomes a hot ESG topic, we may see members/stakeholders become more vocal. This might suggest that taking more (and more thoughtful) stewardship action on AI is a greater priority.
Summing up
The potential effects of AI development span operational resilience, market risk, labour markets, covenant strength and stewardship priorities. Asset owners should carve out time in their meeting agendas for training on AI and its implications, as well as to engage with asset managers or insurers for pension schemes looking to transfer risk. In addition, there are several other AI-related tasks to consider, such as exploring the investment implications, conducting scenario analysis, introducing ongoing monitoring and improving governance around how AI is used in your operations.
|
