AI Evaluation Needs Repeatable Methods, Not Just Deeper Access
Research and evaluator frameworks distinguish repeatable measurements, documented results and the access needed to examine hidden failure modes.
Share
Posts by tag
Safety
11 posts
ChatGPT for Teens Adds Guardrails Years After Young Users Arrived
OpenAI has introduced a teen-specific ChatGPT experience with study tools, family controls and stronger defaults, but key questions about enforcement remain.
Share
China’s AI Companion Rules Target Emotional Dependence, Not All Agents
Beijing requires anti-addiction, minor protection and distress intervention for anthropomorphic services, prompting major platforms to shut companion features.
Share
Claude Sonnet 5 Arrives as Anthropic Restores Its Frontier Models
A new safety classifier ended an 18-day Fable and Mythos shutdown while Sonnet 5 targets capable agentic work at lower cost.
Share
Autonomous Machines Force AI Governance Into the Physical World
Singapore’s agentic AI framework shows why robots need staged deployment, continuous monitoring and clear human accountability.
Share
Physical AI Turns Governance Into a Real-World Safety System
As AI moves into robots and industrial equipment, governance must connect model controls with mechanical limits, audit trails and safe escalation.
Share
Stanford Warns AI Safety Benchmarks Are Falling Behind
Stanford's 2026 AI Index finds inconsistent safety reporting, rising incidents and declining organisational confidence in incident response.
Share
Glia Makes AI Safety Part of the Banking Service Proposition
Glia’s award-winning banking platform combines high interaction automation with contractual commitments around hallucinations and prompt injection.
Share
ADLINK and Noble Machines Target Robots at Dangerous Work
ADLINK and Noble Machines are combining rugged edge hardware with whole-body autonomy for general-purpose robots in hazardous industrial settings.
Share
Hitachi Brings Industrial Knowledge to the Physical AI Race
Hitachi is combining engineering expertise, reusable software and strict safety controls to deploy physical AI in factories and railway systems.
Share