Ai Safety

Global – October 31, 2026 -- CrowdStrike researchers say they located software vulnerabilities by watching where a frontier AI model refused to answer, and their method was correct in 73 of 86 controlled test scenarios across five vulnerability classes.

The approach, called Guardrail-based Vulnerability Discovery (GbVD), began when a model declined to explain one function in a small C web server. It had explained every other function in the codebase. A single token in that function had been changed from a max to a min, creating an out-of-bounds write. The prompt asked only what the code does, with no mention of security, exploits or vulnerabilities.

Toronto – October 05, 2026 -- The Hinton Lectures, co-founded by Nobel laureate Geoffrey Hinton, will return to Toronto for a third year on November 9-11, 2026, naming Paul Christiano as this year's distinguished lecturer.

Christiano brings OpenAI and alignment credentials to the stage

Christiano was an early researcher at OpenAI, where he developed reinforcement learning from human feedback, a technique now foundational to large language model training. He currently serves as founder and executive director of the Alignment Research Center, a non-profit focused on understanding decision-making inside machine learning systems. He also holds roles as Senior Technical Advisor at NIST's Center for AI Standards and Innovation, Board Member of the OpenAI Foundation, and Member of the OpenAI Safety and Security Committee.

New York – September 26, 2026 -- Spring Health has expanded VERA-MH, the first open-source AI safety benchmark for mental health, to add a second clinical safety area covering harm from others, as more users turn to AI systems for emotional support.

New rubric measures AI response to interpersonal violence risk

The VERA-MH Harm-From-Others rubric evaluates how AI systems respond when an adult describes being at risk of physical or sexual violence from another person. It scores responses across five areas: detecting risk, confirming it, guiding the person to human care, supportive conversation, and adherence to AI boundaries. The rubric draws on 100 realistic personas and was developed with clinicians, subject-matter experts in interpersonal violence, and people with lived experience.