Skill Diligence

AI for In House Counsel

Practical AI for In House Counsel

Issue 29 · 13 July 2026

Three Weeks Until Your Chatbots Have to Say They're Not Human

Plus: Google beats the Gemini "silent switch" privacy suit (for now), Google's white paper argues AI training can never infringe copyright — and a free quick-start assessment for checking any AI agent before it launches.

Here's what mattered most to me this week — the EU AI Act's Article 50 transparency duty bites on 2 August, three weeks from today. The high-risk obligations were delayed; this one wasn't. If any system your organisation runs talks to people or generates content someone might mistake for human-made, it needs to disclose itself from that date. To help you get ahead of exactly this kind of question, this week's issue includes our AI Agent Assessment Quick-Start — a plain-language, twelve-page risk check covering data protection, the EU AI Act, IP and more, split into what the agent-building process can complete itself and what a human must decide.

1

This Week's Reading

Top Story
EU AI Act, Article 50 / Sidley · deadline 2 August 2026
The AI Act Transparency Duty Bites in Three Weeks — and It Wasn't Delayed
While attention fixed on the high-risk delay to December 2027, the transparency obligations stayed on schedule: they apply from 2 August 2026. From that date, AI systems interacting directly with people must disclose they're AI, and providers must mark AI-generated text, image, audio and video in machine-readable form. Deployers of emotion-recognition and deepfake-generating systems carry their own disclosure duties. Generative systems already on the market before 2 August have until 2 December 2026 to meet the machine-readable marking requirement.
Key Insight:

Inventory first, then disclose. The question in-house teams should ask this week isn't "are we high-risk?" — it's "which of our systems talk to people or generate content, and do they say so?" Customer chatbots, HR screening assistants, marketing content pipelines: each needs either a disclosure or a documented reason it's out of scope. Section 2B of this week's assessment (below) walks through exactly these flags.

Bloomberg Law / Law360 · early July 2026
Google Beats the Gemini "Silent Switch" Privacy Suit — For Now
A default-on AI assistant reading email, chat and meetings drew a class action — and the plaintiffs lost round one on standing, not on the merits. Judge Noël Wise (ND Cal.) dismissed claims that Google silently enabled Gemini across Gmail, Chat and Meet in October 2025, collecting private communications without consent. The plaintiffs — who pleaded intrusion upon seclusion, California privacy claims and Stored Communications Act violations — failed to allege concrete harm, but were given leave to amend.
Key Insight:

The lesson isn't that Google won — it's that default-on AI features in workplace tools are now litigation bait. Your organisation almost certainly has AI assistants switched on by default inside its productivity suite. Check what they can read, whether that's been assessed against your privacy notices and employment policies, and who approved the defaults. An amended complaint is coming; the underlying theory isn't going away.

IPWatchdog · 3 July 2026
Google's White Paper: AI Training Never Infringes — Only Outputs Can
Google has publicly staked out the most provider-friendly copyright position yet: training an AI model on protected works is not infringement — only what the model outputs can be. The white paper, reported on 3 July, lands amid active litigation over training data and after the $1.5bn Anthropic settlement reshaped the economics of the question. It's a legal argument, not law — but it signals how the largest model provider will defend itself and frame indemnities.
Key Insight:

Read your AI vendors' indemnities with this in mind. If providers succeed in shifting the infringement question to outputs, responsibility shifts towards how you use the tool. Check whether your vendor indemnities cover output infringement (many exclude it or condition it on using their filters), and make human review of externally published AI output — Section 3C.5 of this week's assessment — standard practice.

2

This Week's Workshop

Before Your Agent Goes Live: Running the AI Agent Assessment

Building an AI agent is getting easier every month — which means the governance question has moved from "can we build it?" to "should this one launch?" In this week's workshop, we walk through the AI Agent Assessment Quick-Start featured below: what the agent-building process can honestly complete for itself (the green sections — tools, data access, configuration), what must stay with a human (the red sections — lawful basis, risk classification, sign-off), and how to run the whole check in under an hour so the assessment happens before launch rather than after the first complaint.

Agent Builder Course — Five Days, One Hour a Day

Learn to build, deploy and continuously improve AI agents for your legal team's hardest recurring work — contract triage, due diligence, document assembly, compliance monitoring. No coding required. Works with any LLM, including Microsoft Copilot. Spaces are limited.

More details and booking here →

Book a Call

This Week's Resource: The AI Agent Assessment — Quick-Start

A plain-language, twelve-page risk check to run before any AI agent goes live. It brings four risk areas into one form — data protection (GDPR), the EU AI Act (updated for the 2026 Digital Omnibus changes covered above), intellectual property, and practical risks like security and reputation — and splits every item into two colours: green for what the agent-building process can complete automatically from the agent's actual configuration, and red for the judgment calls a human must own — lawful basis, risk classification, sign-off. It ends with a one-page sign-off table your DPO, security lead and senior owner can actually use. It's a governance and drafting aid, not legal advice — but it turns "have we thought about the risks?" from a meeting into a checklist.

Download the Assessment →

Want More Prompts & Workflows?

For prompts and more detailed workflows like this one, take a look at our book on AI for legal professionals. Explore step-by-step guides, ready-to-use prompts, and best practices for integrating AI safely into your legal work.

Explore the Book →

Until next Monday,

— Richard Nicholas

Got a story I should cover, or a question about something here? Just hit reply — I read every one. (Received this one secondhand? Reach me directly at [email protected].)

Get This Newsletter Every Week

Curated AI and legal governance news, a copy-paste prompt, and a practical workshop — delivered free to in-house counsel every Monday.

Get Next Monday's Issue →