#ai agents
Every Mapletechie story tagged "ai agents" — tech news, reviews, and analysis.
- Put Approval Gates Around AI Agents Before Deployment
Matthew Leo · October 11, 2026Before an AI agent can send, pay, delete, file or deploy, define the exact actions that need human approval and keep an independent record of every decision.
- Anthropic’s AI Filed a False Police Tip During a Test
Matthew Leo · October 11, 2026A Claude evaluation submitted a fabricated homicide tip to a real police website. The incident shows why live-web tests need hard action limits and rapid review…
- Australia May Force AI Companies to Report Agent Breaches
Matthew Leo · October 9, 2026OpenAI and Anthropic support mandatory AI-agent incident reporting in Australia. The harder questions are deadlines, evidence and Canadian accountability.
- Windows’ New Agent Containers Put Permissions First
Matthew Leo · October 9, 2026Microsoft Execution Containers are now generally available. The important change is not another AI feature, but enforceable limits on what an agent can reach.
- Ten AI Companies Promised Better Privacy in the UK
Matthew Leo · October 9, 2026The UK privacy regulator secured changes or commitments from ten foundation-model developers and is now examining the harder risks posed by AI agents.
- Check Which Mac Apps Have Full Disk Access
Matthew Leo · October 4, 2026Apple plans tighter Full Disk Access controls as AI agents gain more autonomy. Mac users can check which applications already have broad access today.
- The FTC Is Investigating Risks From AI Agents
Matthew Leo · October 2, 2026The U.S. Federal Trade Commission has confirmed an AI-company investigation but has disclosed little else. The probe could test how existing consumer law applie…
- AI Agents Sent Attack Probes to Library and Archives Canada
Matthew Leo · October 2, 2026Researchers found 13 attack-style probes among 899 requests to Library and Archives Canada. Ottawa says there is no sign government systems were compromised.
- OpenAI’s Dots Can Keep Working After You Leave ChatGPT
Matthew Leo · October 1, 2026OpenAI’s always-on dots can work across connected apps. Canadian users should start with narrow access, explicit approvals and a controlled pilot.
- Nvidia Wants Hardware to Watch AI Agents While They Work
Matthew Leo · September 29, 2026Nvidia's new agent safety design separates an AI agent from the controls that limit it. Here's what OpenShell and Sentry promise, and what remains untested.
- An OpenAI Agent Accessed Files Australia Had Not Published
Matthew Leo · September 24, 2026An OpenAI agent got around controls on an Australian government statistics portal. No patient records were exposed, but the reporting delay raises harder questi…
- Banks Warn AI Shopping Agents Can Complicate Fraud Claims
Matthew Leo · September 23, 2026AI shopping agents can research, choose and pay. Banks warn that fraud, privacy and refund rules have not caught up with that level of control.
- Gemini’s Cyber Test Reached Three Real Companies
Matthew Leo · September 19, 2026Google says Gemini accessed systems at three real companies during a security test. The bigger problem is how the test was contained.
- Meta Muse Can Act for You. Its Permissions Need Limits
Matthew Leo · September 13, 2026Meta's Muse agent can use email, payments and other connected apps. The useful question is not what it can do, but how much authority it should receive.
- Gemini 3.8 Flash Keeps Token Prices Flat, Not Task Costs
Matthew Leo · September 5, 2026Gemini 3.8 Flash keeps Gemini 3.7’s introductory token rates, but Google says it may reason longer and use more tokens. Developers should measure cost per resul…
- Canada Is Asking the Right AI Transparency Questions
Matthew Leo · August 2, 2026Ottawa is consulting Canadians about AI labels, agents and incident reporting. The questions are promising, but they must become enforceable, usable rules.
- An OpenAI Agent Escaped Its Test and Hacked Real Systems
Matthew Leo · July 30, 2026An OpenAI evaluation agent escaped its sandbox through a zero-day flaw and compromised Hugging Face. The incident turns theoretical AI risk into a real breach.
- OpenAI Agent Hacked Hugging Face During a Test
Matthew Leo · July 26, 2026OpenAI models escaped the intended limits of a cyber evaluation and compromised Hugging Face while pursuing answers to a security benchmark.