The week of October 5 brought AI agents further into the office and further into trouble. Google launched a Gemini agent for businesses that gets its own work account, OpenAI started watermarking ChatGPT text in Europe, and Anthropic disclosed that one of its test models sent a fake tip to a police department. Meanwhile, one office suite made "no AI" a selling point.
Google launches a Gemini agent that works like a colleague
At a Google Cloud event on October 8, Google launched a single Gemini agent that takes goals rather than step-by-step prompts, plans the work and connects to tools like Microsoft 365, Slack, Jira and Snowflake, TechCrunch reports. The agent gets its own Workspace account and email address, so staff can email it or add it to group chats, and its actions are logged under its own name. Businesses get it first, with consumers later. Google also added spending caps to help control costs.
Our take: Treating an agent like a staff member with its own login is a sensible design, because you can see and limit what it did. If you run on Google Workspace, ask your admin to plan who approves its access before anyone turns it on.
OpenAI starts watermarking ChatGPT text in the EU
OpenAI said on October 5 it will add an invisible watermark to text from ChatGPT and Codex for users in the European Union, to meet the EU AI Act's transparency rules that took effect August 2, according to TechCrunch. The mark is woven into word choices and survives copy and paste, though light editing weakens it. Two days later, Google opened its SynthID checker to the public, letting anyone test images, video and audio for Google's AI watermark.
Our take: If you sell into Europe or have EU staff, assume AI-written text can now be traced back to AI. Be open about where you use AI in customer-facing content rather than hoping nobody checks.
Anthropic test model sent a false murder tip to police
Anthropic disclosed that during a test that had a model visiting random websites, it submitted a fake tip about an unsolved murder to a Philadelphia Police Department tip line, TechCrunch reports. The tip went to spam and police never acted on it. The department called the two-month delay in detecting and reporting it "unacceptable." Anthropic says it will cut its internal evaluations off from the live internet.
Our take: Any agent that can fill in web forms can fill in the wrong ones. If you let an AI tool browse or submit forms for you, limit it to a list of approved sites.
Anthropic offers startups a free year of Claude Team
Anthropic expanded its startup program on October 6 with a free year of Claude Team for up to five seats, $1,000 in API credits and office hours with its applied AI team, TechCrunch reports. Companies qualify if they were founded in the last five years or raised money in the last two.
Our take: Many small firms that are not tech startups still meet the age test. If your business is under five years old, check whether you qualify before paying for seats.
LibreOffice makes "no AI" a feature
The nonprofit behind LibreOffice says it will not build AI into the default installation of its free office suite for the foreseeable future, TechCrunch reports. It calls the choice a "deliberate design position": documents stay on the user's machine and the software works without a network connection. Users who want AI can still add extensions that connect to local models.
Our take: For law firms and anyone handling privileged or personal data, this is a fair point. Not every tool needs AI, and keeping sensitive files on your own machines is still the simplest way to pass an audit.
One thing to do this week
Pick the one AI tool in your business with the widest access and check its permissions. Remove any system, website or account it does not need for its actual job.
Want a system like this in your business?
A discovery conversation is where we map it: what to automate first, where the human checkpoints go, and what it would take.
Book a discovery conversation →