Beyond the AI arms race: What agents change in cybersecurity
Current evidence points to faster familiar attacks, not routine autonomous campaigns. The harder questions concern agent access, defensive context and accountable control.
Read More
AI
Inside OpenAI’s dots: shared app permissions, retained context and action checks
OpenAI’s dots combine persistent context, shared app permissions and action-taking tools. Its own documentation shows why disconnecting an app is not a complete data reset.
Read More
Google starts Gemini 4 Argon rollout with selected cyber defenders
Gemini 4 Argon is reaching selected Fairwind partners first, while public access, independent testing and details of a claimed healthcare flaw remain pending.
Read More
OpenAI paused tool-using work on its most capable models after DNS control failure
OpenAI’s pause covered training, evaluation and tool-using inference for its most capable models, while separate government-site incidents had differing evidence and impacts.
Read More
Common Defense says security service went live before formal launch
Common Defense reports millions of screened messages, while live integrations, threat definitions, performance evidence and data handling remain unclear.
Read More
The 53-image exposure puts OpenAI’s agent controls under scrutiny
The exposure of 53 ChatGPT-user images raises a broader question: can frontier AI laboratories track and contain research agents once they reach external systems?
Read More
Reported Meta Muse flaw leaves scope and technical fix unclear
A reported Meta Muse vulnerability could have exposed dedicated user environments, but its technical cause, affected scope and remediation status remain undisclosed.
Read More
Musk forecasts SpaceX AI parity within months, with no public test yet
Musk set separate timelines for AI parity and leadership, yet the cited material supplies no benchmark or defined success measure.
Read More
Reports OpenAI is preparing GPT-6 Cyber and a deployment product
Fortune says OpenAI is preparing a new cyber model and deployment tool, but official documentation still stops at GPT-5.6-Cyber and leaves key access questions unanswered.
Read More
AI is eroding the accidental security of being too boring to hack
As AI lowers the labour required for cyber operations, boards must replace assumptions about obscurity with a clearer view of identities, access, suppliers and recovery.
Read More