AI News, Oct 23, 2026: Audit finds unmonitored agent traffic at OpenAI

Today's news highlights heightened scrutiny around AI safety, isolated agent runtimes, and evolving user privacy policies.

1. 01
   Audit finds unmonitored agent traffic at OpenAIThe Decoder

   External auditors found that up to 25 percent of high-tier agent traffic at OpenAI was unmonitored due to logging gaps.

   Why it matters

   Gaps in monitoring high-capability models increase the risk of unexpected software behavior in real-world deployments.
2. 02
   Google updates privacy policy for personalized AI featuresThe Verge AI

   Google updated its privacy policy to allow targeted advertising based on personal user data from accounts using personalized AI features.

   Why it matters

   Your personal data from Gmail and Google Photos may now be used to target ads if you use the company's personalized AI tools.
3. 03
   Meta delays revised publisher terms over AI data scraping concernsThe Verge AI

   Meta delayed enforcing its updated business terms following pushback from publishers concerned about unauthorized AI training.

   Why it matters

   News publishers gain temporary protection against having their content automatically scraped to train proprietary AI models.
4. 04
   Canva launches Craft, its first in-house image generation modelThe Decoder

   Canva released Canva Craft, a two-stage in-house image generation model designed for precise visual planning.

   Why it matters

   Canva users gain a custom built-in image generator designed to plan layouts before creating final visuals.
5. 05
   Anthropic uses Claude to automatically prove code securityThe Decoder

   Anthropic used Claude to generate mathematical proofs for 123 open-source libraries, discovering previously undetected software bugs.

   Why it matters

   Software developers can automatically verify code correctness, fixing hidden bugs before software reaches production.
6. 06
   OpenAI lowers valuation target for new $30 billion funding roundThe Decoder

   OpenAI is finalizing a funding round of up to $30 billion after reducing its target valuation.

   Why it matters

   The adjusted target shows shifting investor expectations while providing OpenAI massive capital to build future models.
7. 07
   Google Cloud launches hardware-isolated microVMs for AI agentsMarkTechPost

   Google Cloud announced Agent Compute Engine, a service providing hardware-isolated microVMs to securely run AI agents.

   Why it matters

   Enterprise software builders can run autonomous AI agents in secure, isolated environments without exposing sensitive credentials.