25/03/2026
The International AI Safety Report 2026 has just been released, and it serves as a sobering reminder of the pace of change in our digital landscape.
For those of us in governance roles across Aotearoa New Zealand, this second global scientific assessment provides essential data to inform our strategic governance and risk oversight.
General-Purpose Artificial Intelligence (GPAI)—defined as Artificial Intelligence (AI) models capable of performing a wide variety of tasks—has seen significant capability gains in the last year.
However, for the professional director, the report highlights that these advances come with a "jagged" performance profile. While systems now match expert-level performance in complex reasoning, they still fail at simple tasks or recover poorly from basic errors.
Here are my three key takeaways for company governors:
Mind the Evaluation Gap: There is a growing disconnect between how AI performs in lab tests and how it behaves in the real world. Boards should not rely solely on vendor-supplied benchmarks. We must ensure our management teams are conducting robust, context-specific testing within our own business environments.
Address Automation Bias: As these tools become more integrated into our daily mahi, there is a significant risk of staff trusting AI outputs without enough scrutiny. Maintaining human autonomy and critical thinking is a matter of guardianship over our intellectual capital.
Adopt Defence-in-Depth: No single safeguard is perfectly reliable. The report advocates for a "defence-in-depth" approach—layering technical filters, organisational policies, and continuous monitoring to manage risks like cyberattacks and data inaccuracies.
As we look toward the long-term horizon of 2030, the trajectory of this technology remains uncertain. Our role as directors is to ensure our organisations remain resilient, regardless of whether AI progress plateaus or accelerates.
I encourage all fellow governors to review the executive summary to understand how these developing issues might impact your specific industry.
The second International AI Safety Report, published in February 2026, is the next iteration of the comprehensive review of latest scientific research on the capabilities and risks of general-purpose AI systems. Led by Turing Award winner Yoshua Bengio and authored by over 100 AI experts, the report...