Five fresh stories for tech professionals: robo-firing limits, Reddit’s data lockdown, tiny decision models, Copilot’s desktop control and a new agent liability bill.
It was a policy-heavy few days in AI. Below is a quick-scan terminal, then the full story behind each item. Tap through the tabs:
[2026-10-01 policy] California signs SB 947
California became the first state to bar employers from relying only on automated systems to fire or discipline workers.
Why it matters: Any HR or workflow tool you build needs a human review step and an audit trail.
[2026-09-30 data] Reddit closes RSS and public API
RSS ends November 13 and the public API closes by March 2027. Developers must register by January 12.
Why it matters: Bots, research tools and AI products that read Reddit will need commercial access.
[2026-10-01 models] decision-model wave grows
Cloudflare released Clef and Perplexity launched a Decisions API, both returning typed answers with probabilities instead of prose.
Why it matters: Routing and guardrail steps in agent pipelines can use small, cheap, auditable models.
[2026-10-01 tools] Copilot computer use preview
GitHub Copilot CLI and app can now click, type and scroll in desktop apps on macOS and Windows, after you approve access.
Why it matters: Legacy GUI-only tools can finally join an automation workflow, with admins able to disable it.
[2026-10-01 law] AI Agent Accountability Act
Senators Hawley and Murphy announced a bipartisan bill creating liability for AI-enabled hacking.
Why it matters: Teams running autonomous agents should document safeguards and access limits now.
1. California bans AI-only firings
Governor Newsom signed SB 947, the so-called No Robo Bosses Act. California is the first state to prohibit employers from relying solely on automated systems to fire or discipline workers. The law requires human corroboration, written notice and access to the underlying data, and takes effect July 1, 2027.
Why it matters: If you build HR, scheduling or performance tooling, a human-in-the-loop step is now a legal design requirement in the largest US state economy. Log every automated recommendation so a reviewer can show what they saw. Source: Quartz.
2. Reddit shuts RSS and the public API
Reddit says it will end RSS feeds on November 13 and close its public API by March 2027, citing scraping and automated abuse. Third-party developers must register by January 12, and AI products will need commercial licensing to keep using the data.
Why it matters: Monitoring bots, social listening dashboards and research pipelines that quietly depend on Reddit data have a deadline. Audit your integrations now and plan for licensed or alternative sources. Source: TechCrunch.
3. Update: the decision-model wave spreads
Earlier decision-style APIs were covered in previous roundups. The new development is that more vendors joined on October 1. Cloudflare released Clef, an open model for scoring allowed choices, and Perplexity launched a Decisions API that answers yes/no and multiple-choice questions with probabilities rather than paragraphs.
Why it matters: Agent pipelines often waste big models on small calls such as routing a ticket or approving a step. Typed, low-latency decision models are easier to test and audit. Source: Perplexity docs and Dealroom.
4. GitHub Copilot gets computer use
GitHub put computer use into public preview for the Copilot CLI and app on macOS and Windows. Copilot can read accessible app content, click controls, type, scroll and drag, but it asks for approval first, and organizations can disable the feature.
Why it matters: Plenty of enterprise tools have no API or MCP server. This gives automation a path into them, but treat it like any privileged operator: scope it, review the approvals and keep it off production consoles. Source: GitHub Changelog.
5. Senators unveil the AI Agent Accountability Act
Senators Josh Hawley and Chris Murphy announced bipartisan legislation that would create civil and criminal liability for AI-enabled hacking, depending on knowledge, recklessness and the safeguards in place.
Why it matters: Security and platform teams that deploy autonomous agents should be ready to show access limits, logging and kill switches. Safeguards are the proposed dividing line. For hands-on practice, see our courses on DevOps, security and AI. Source: Sen. Hawley.
Roundup checks
This edition was checked against the previous AI roundups, including Sep 29, Sep 30 and Oct 1, to avoid repeating stories.
