AI is moving deeper into critical infrastructure, private communication, and Windows itself

Six major stories from Aug. 19–20: Siemens PLC attacks, ChatGPT in iMessage, private safety processing, Stripe’s OpenRouter deal, and NPU tracking 👇

1. US agencies warn of active AI-generated exploits targeting Siemens S7 PLCs in critical infrastructure Sources: CISA/NSA/FBI/DOE/EPA advisory · SecurityWeek

Joint advisory (Aug 19) details threat actors using AI-assisted scripts (with open-source snap7 libraries) to target internet-exposed Siemens S7 Series PLCs for reconnaissance, access, and potential disruption across energy, water, manufacturing, and other sectors. Why it matters: Demonstrates AI lowering the bar for ICS/OT attacks, accelerating real-world critical infrastructure risks.

2. ChatGPT gains Apple Messages plugin that can read, draft, and send iMessages Sources: TechCrunch · MacRumors

OpenAI launched a macOS plugin (Aug 20) letting ChatGPT Work/Codex search, analyze, draft, and (with approval) send iMessage/SMS/RCS messages locally on Apple Silicon Macs. Why it matters: Advances agentic desktop automation into personal messaging, raising privacy and permission questions.

3. OpenAI previews Private Safety Processing for zero data retention on frontier models Sources: OpenAI · Axios

OpenAI announced (Aug 19) Zero Data Retention for eligible API customers plus Private Safety Processing that detects misuse patterns across interactions without staff accessing prompts/responses; testing underway, broader rollout September. Why it matters: Resolves enterprise tension between safety monitoring and strict privacy for agentic/enterprise deployments.

4. Stripe acquires OpenRouter in ~$7.5–8B deal Sources: OpenRouter · New York Times

Stripe agreed (Aug 19) to acquire OpenRouter, the multi-model routing/marketplace processing 10+ trillion tokens daily across 400+ models for 10M+ users; product and integrations unchanged, close expected soon. Why it matters: Consolidates token routing, metering, and payments into a single infrastructure layer for AI-native SaaS.

5. Anthropic plans to let enterprises keep retention data on their own cloud Sources: Reuters · The Decoder

Anthropic plans (reported Aug 20) to revise its 30-day retention requirement for Mythos/Fable models so enterprises can store the data on their own infrastructure while still enabling safety monitoring; rollout later this year after work with 100+ customers. Why it matters: Responds to enterprise pushback and competitive pressure from OpenAI’s zero-retention approach.

6. Microsoft adds per-process AI/NPU monitoring to Windows 11 Task Manager Sources: The Register · Microsoft Tech Community

Microsoft updated Task Manager (announced ~Aug 19–20) with optional NPU, NPU Engine, dedicated/shared NPU memory columns and GPU neural engine visibility on supported devices. Why it matters: Gives developers and IT real-time insight into local AI workloads as on-device inference grows.

GET BONUS ON BLOFIN | JOIN CHAT | X