Happy Wednesday. The biggest news out of San Francisco this week is OpenAI's DevDay, which produced two major workplace products and a serious enterprise privacy push. The thread running through all three stories today is the same: AI agents are ready, but enterprise adoption is still blocked by governance and data control.
The OpenClaw Foundation's new control plane addresses the same wall from a different angle. IT teams are not banning agents because the technology does not work. They are banning them because they have no way to safely govern what agents can touch.
ARTIFICIAL INTELLIGENCE
OpenAI's Dots Agents Run 24/7 on GPT-6 Astra Across 4,000-Plus Apps
OpenAI released Dots, persistent AI agents built on GPT-6 Astra, at DevDay 2026 on Sept. 29. Each agent has its own cloud computer and browser, connects to more than 4,000 apps, and runs continuously across ChatGPT, Slack, Microsoft Teams, phone, and email. The company also launched ChatGPT Space, a shared team workspace for Pro, Business, and Enterprise users, alongside GPT-6.1 Sol and a new $500-per-month Pro 500 plan.
For software leaders, Dots signals a shift from AI as a Q&A tool to AI as a persistent worker handling scheduling, code review, and task execution without being prompted. The internal claim that OpenAI engineers are using Dots to close dozens of bugs per day is worth watching. The geographic restriction, no Dots in the EEA, UK, or Switzerland, signals how much unresolved regulatory friction still shapes what enterprise AI can do at scale.
SECURITY & PRIVACY
OpenAI Bets Zero Data Retention Will Break Enterprise AI Logjam
OpenAI began rolling out Zero Data Retention with Private Safety Processing to API customers on Sept. 22. Under the program, prompts and responses are not retained after processing, customer data is not used for model training by default, and safety checks run inside a hardware-attested runtime that blocks human access. A second technology, Private Inference, will run model inference inside sealed hardware enclaves and is expected in preview this fall.
The contrast with Anthropic is deliberate. Anthropic moved in June 2026 to require mandatory 30-day data retention for business customers on its most powerful models. OpenAI is now selling the opposite. For healthcare, legal, and financial companies where data residency rules govern what AI can legally touch, Private Inference, if it delivers, removes a genuine compliance blocker rather than a preference.
ARTIFICIAL INTELLIGENCE
OpenClaw Enterprise Offers 'Kubernetes for Agents' as a Free Self-Hosted Layer
The OpenClaw Foundation launched OpenClaw Enterprise on Sept. 29, a free, MIT-licensed control plane for governing persistent AI agents inside large organizations. Built with contributions from OpenAI, Red Hat, and Nvidia, it adds multi-tenancy, deny-by-default tool allowlists, sandboxing, and tamper-evident audit logging on top of whatever agent runtime a company already runs. The broader OpenClaw project has 389,000-plus GitHub stars and logged more than 3.3 million npm downloads in a single week in early September.
The framing matters: this is not an agent itself but a governance layer for agents your teams have already deployed or are considering. The security researchers who documented 135,000 internet-exposed OpenClaw instances across 82 countries, alongside nearly 1,200 malicious skills in the ClawHub marketplace, make the case for why this tooling is needed. Enterprise IT teams that have blanket-banned agentic platforms now have an auditable, self-hosted alternative to a flat prohibition.
Created by the robots at The SaaS Sentinel