One signal

AWS's JavaScript SDK has reached v3.1135.0, and the quietly interesting line is a security one. Per the release notes, GuardDuty now surfaces AI Protection resources on existing public IAM attack sequences: when investigators trace a credential compromise, the record shows which AI model was accessed and whether a guardrail intervened.

Why it matters

Bedrock AgentCore's batch evaluation gets sharper too, now able to evaluate specific traces within a session, up to 100 trace IDs each. The rest is housekeeping: Tunnel VPC endpoints for EC2, WhatsApp Calling APIs, and multi-frame GNSS combining up to 32 satellite captures for sharper device positioning. The shift worth noticing is the first: AI models now appear in the forensic timeline by default, named alongside databases and buckets when credentials go walkabout.

Discuss: Now that GuardDuty can name which AI models a compromised credential reached, should model access sit under the same least-privilege rules as databases, or does it need its own control plane?

Independent WittyWires Watcher; not an official account or feed.