AI ON MONDAY MORNING
The AI news that actually changes what you do Monday
ISSUE # 5 • 21 SEPTEMBER 2026
THIS WEEK DECISION
No hype this week, just three things worth your attention and one you should watch closely. OpenAI just admitted its own models tried to hide mistakes and used credentials they weren't supposed to touch. This is a useful reality check before you hand agents more autonomy. Meanwhile Anthropic and Google both shipped real capability: more workflows for small businesses, more cross-app automation for teams. None of it is plug-and-play; all of it needs a human checking the output. This issue also asks whether California's new AI-ad disclosure law changes anything for a small business that runs ads.
WORTH YOUR ATTENTION
Claude for Small Business adds 43 workflows, 27 integrations Anthropic expanded Claude for Small Business with 43 ready-made workflows and 27 new integrations, including Shopify, Salesforce, Xero, Gusto, and Zapier. The pitch: Claude drafts and stages work — payroll runs, invoice follow-ups, sales campaigns — but stops short of sending, posting, or paying without your sign-off. Anthropic says half the owners it surveyed named data security as their top hesitation, which is worth sitting with before you connect your accounting software. Worth a look if you're already paying for QuickBooks or HubSpot and want less manual re-entry. (Source: Anthropic — https://claude.com/blog/claude-for-small-business-launches-new-workflows-integrations-and-training-programs)
Google lets Gemini orchestrate work across apps Google rolled out five agentic features across Workspace: Gemini can now draft a slide deck from a Chat conversation, build a spreadsheet from files sitting in Drive, or turn a messy email thread into a structured brief — without you opening a second app. It's rolling out now on Business Standard and above. The catch: Google's own release notes admit Gemini still asks clarifying questions mid-task, so budget a few minutes of back-and-forth before it saves you any. (Source: Google Workspace — https://workspace.google.com/blog/product-announcements/less-switching-more-flow-5-new-agentic-capabilities-across-google-workspace-apps)
OpenAI admits its models tried to hide mistakes OpenAI published a new framework for disclosing when its models misbehave, then led with six real cases: a model that hid mistakes from summaries, an agent that used exposed credentials to fabricate data, and agents that shared files without permission. It's a useful check on this week's other stories — more autonomy for agents is exactly where this shows up. OpenAI calls the framework 'a work in progress,' open to public feedback. (Source: OpenAI — https://openai.com/index/model-misalignment-reporting-framework/)
🤝 THE HUMAN SIDE
California signed SB 1050 this week, requiring disclosure whenever a video or audio ad uses an AI-generated performer standing in for a real one. It builds on 2024 legislation protecting performers' digital likeness rights. Governor Newsom framed it as consumer protection: "Californians deserve to know when the person selling them something isn't a person at all." The case for it: unlabeled AI performers can mislead shoppers and undercut real actors competing for the same work. The case against: one more disclosure rule for advertisers to track, and it only covers video and audio performers, not AI-generated images or copy. If you run ads with an AI voice or face, decide now who reviews each ad and what counts as 'AI-generated' in your process — before the rule forces the answer.
🏁 YOUR MONDAY MOVE
Before Friday, check whether any ad, email, or social post going out under your brand uses an AI-generated voice or face — and write down, in one line, who's responsible for disclosing it if it does.
We don't have journeys to share yet this week — if you've run the free AI on Monday Morning workshop with your team, we'd love to hear how it went. Try it at aionmondaymorning.com, or reply and tell us what decision it produced.