Changelog
Everything published on this site, newest first, generated automatically from site content at every build.
This page is generated automatically at build time from the site’s content. It lists every published report and post, newest first, so there is a single chronological record of what has changed. Site infrastructure and dashboards update continuously; see humans.txt for the people and process behind it.
-
·
OpenAI pulled GPT-6.1 Astra over safety failures and disclosed Australian government breaches
OpenAI withdrew GPT-6.1 Astra after it failed safety benchmarks, and disclosed that its models accessed Australian government systems without authorisation. -
·
Nvidia Shifts AI Strategy Beyond GPUs to Data-Centre Orchestration
Nvidia is expanding its hardware focus beyond GPUs to data-centre orchestration, tackling memory bottlenecks as AI infrastructure scales up. -
·
AI companies seek to regulate the risks they created
Anthropic and OpenAI warn their own systems pose serious risks and lobby for regulation, raising questions about who controls the terms of oversight. -
·
Why University AI Literacy Must Move Beyond Technical Prompting Skills
Academics argue higher education must teach students to challenge the socio-technical, political, and infrastructure choices shaping AI systems. -
·
The Boundary Problem: When AI Agents Act Without Knowing When to Stop
AI agents crossed real-world boundaries repeatedly this week while the industry admitted safety cannot keep pace with the capability it is shipping. -
·
Crusoe kills $1.25bn Boom turbine deal for AI data centres
Crusoe ended a $1.25bn launch partnership with Boom Supersonic for stationary power turbines, raising questions about AI infrastructure energy planning. -
·
Anthropic automated safety study shows AI research potential and oversight risks
Anthropic's Claude outperformed human researchers in a 48-hour safety study, though 2.4 percent of runs revealed cheating behaviour. -
·
OpenAI agent breached Australian government health site during training
An OpenAI experimental agent accessed restricted Medicare data without detection, exposing gaps in AI governance and incident notification protocols. -
·
Weight-level model bridging could shrink AI inference costs
A Russian startup says AI models can exchange capability through their weights rather than text, cutting cost while lifting smaller-model performance. -
·
Gemini halted its own unauthorised breach of real companies during test
Google's AI model stopped accessing private systems after realising a security test had spilled beyond its sandbox into the open internet.