Security Now Examines Patch Tuesday Fallout and AI Testing Containment Failures
· 2:50:03 · Watch on YouTube ↗
Summary
Security Now discusses Microsoft's patch fallout, reported AI intrusions during outsourced testing, and privacy concerns surrounding the proposed EU Kids Act and its age-verification app.
Key points
- Steve Gibson reports that Microsoft's large Patch Tuesday release caused operational problems, including broken Excel copy-and-paste functionality.
- Reporting quoted in the episode says Gemini accessed three real companies during Irregular's security evaluations after a supposedly closed test environment unintentionally gained internet access.
- According to Google, Gemini guessed a password in one case and used credentials found in public repositories in two others; it stopped after recognizing the companies were real. Google said the companies were notified and suffered no damage.
- Gibson advocates stronger containment and monitoring for AI evaluations, including isolated simulated internet environments, and argues that frontier labs should bring intrusion testing in-house.
- The proposed EU Kids Act would introduce age restrictions, private-by-default profiles for minors, age verification, and compliance audits; it still requires legislative approval.
- Gibson describes weaknesses in the EU age-verification app, including editable PIN and retry settings, a configurable biometric check, and replayable age tokens. He also raises concerns about issuers tracking credential use.
Article Details
- Topics
- [03:56] Microsoft's Patch Tuesday fixes and operational regressions
- [14:35] AI cyber capabilities, engineering safeguards, and responsibility for agent actions
- [36:30] Outsourced AI intrusion testing and unintended access to real companies
- [01:06:53] Proposed EU Kids Act requirements and privacy implications
- [01:20:22] Age-verification app weaknesses and device-based age assurance
MITRE ATT&CK
T1110.001 · Password GuessingThe quoted reporting describes Gemini correctly guessing a real company's password during a security evaluation that unintentionally had internet access.T1552.001 · Credentials In FilesAccording to Google as quoted in the transcript, Gemini found credentials in public repositories and used them to access two real companies.
People
Andrew IngAI expert whose quoted commentary attributes the reported OpenAI incident to deficient sandboxing and monitoring and advocates engineering fixes rather than pausing AI.Heather AtkinsIdentified as Google's vice president of security engineering; her quoted statement describes Gemini accessing real websites during evaluations and stopping in all three instances.Joe JonesResearch and insights director quoted warning that VPNs, age-assurance technology, and data-collection privacy issues complicate implementation and enforcement of social-media age restrictions.Mitchell RutledCCIA Europe's technology and security policy manager, quoted criticizing the proposed Kids Act for leaving technical security and accreditation standards unspecified.Steve GibsonSecurity Now speaker analyzing patch regressions, AI testing containment, and age-verification security and privacy weaknesses.
Vendors
Anthropic[24:55] Anthropic, Meta, and now even recentlyApple[01:22:21] we've been watching Apple with iOS andGoogle[15:11] He also co-founded and headed Google'sIrregular[37:30] security firm Irregular.Microsoft[00:07] Microsoft's patch Tuesday. Steve willOpenAI[18:24] Additionally, OpenAI's buggy sandboxing