Earlier this month Anthropic released news on the new invisible watermarks that will be placed into outputs from Claude in compliance with the EU’s AI Act. And the crowd went wild.
The ability to mark AI-generated content is not a new concept and Google has been doing this with both text and images for some time. The concept of the watermark or sign – whether in text or on an image – gives people both the comfort of knowing when something was generated with the assistance of AI and the concern that they will be judged for using AI to generate.
Around the same time, Varun Anand (the co-founder of Clay), shared the AI Writing Policy that his team adopted. It can be summed up in four main ideas:
These two stories feel heavily related to me. Where a subset of our population is worried about being “caught” using AI, Sophie Alpert (the Clay employee who drafted the policy) has defined what’s human and essential about work (even when using AI).
I’m a big believer in determining where you will and won’t use AI as an organization and then being transparent about this with your staff and communities.
Where are you drawing these lines? How are you feeling about the automatic marking of AI-generated content?
Wishing you a wonderful week!
Amanda
AI News that Matters
In Plain English
Watermark: a hidden signal embedded in AI output that marks it as AI-processed, even when you can’t see or hear it.
C2PA label: an industry-standard tag attached to a file that records how it was created or edited.
GPT Store: OpenAI’s marketplace for custom, user-built versions of ChatGPT.
Claude outputs now carry invisible watermarks
Anthropic published details on embedding hidden watermarks in text, code, and file outputs from new Claude models, a step toward the EU AI Act’s transparency rules. Files also get the C2PA provenance label already used across the industry, and older models will be retrofitted over time. Anthropic notes a watermark shows content was processed by Claude, not necessarily fully AI-authored.
Why it matters
This is part of a broader shift toward built-in AI disclosure. If your organization hasn’t decided how it discloses AI-assisted work, watermarking is making that decision for you by default.
Source: TechCrunch · Anthropic
Try This in 5 Minutes
Write one sentence answering Amanda’s question above: where does your org draw the line on disclosing AI-assisted work? Share it with your team this week, even informally.
Custom GPTs may be nearing the end of the road
Reports say OpenAI has started blocking personal accounts from creating new custom GPTs, a possible early signal that the once heavily-promoted GPT Store feature is being wound down.
Why it matters
If your team built a workflow around a custom GPT, this is a good reminder not to treat any single AI feature as permanent infrastructure — keep a fallback for anything mission-critical.
Source: SQ Magazine
Google now lets you remove its visible AI watermark
Google will now allow users to strip the visible watermark from its AI-generated images, a reversal from its earlier position that visible marking was an important transparency signal.
Why it matters
A visible watermark was one of the easiest ways to spot AI-generated images at a glance. Removing that option makes visual content harder to vet, right as Anthropic and others lean further into disclosure.
Source: TechCrunch
Claude Cowork is now on web, desktop, and mobile
Anthropic confirmed that Claude Cowork, its agentic tool built for non-developer knowledge work, is now accessible across web, desktop, and mobile.
Why it matters
Worth a look if you’ve been curious about handing off multi-step work to AI but don’t have technical staff to set it up — broader access lowers the barrier to just trying it.
Source: Claude Support
ChatGPT projects get more flexible memory settings
OpenAI now lets you switch a project’s memory setting after it’s created — toggling between default memory and project-only memory from Project Settings. Project-only memory keeps that project’s conversations out of memory used elsewhere, and shared projects stay locked to project-only mode.
Why it matters
If you or your team keep client, program, or board work siloed in separate ChatGPT projects, this gives you real control over what does and doesn’t cross between them.
Source: OpenAI Help
Free Resources
Report · Blackbaud Institute
Bridging the AI Effectiveness Gap
AI adoption is growing fast, but effectiveness isn’t keeping pace. A new report from the Blackbaud Institute finds that while 85% of social impact professionals are already using AI at work, only about a third believe their organizations are using it very effectively. The strongest results come from organizations moving past individual experimentation and building real practices around governance, data, infrastructure, and transparency.
A few findings stood out to me: AI maturity is linked to better outcomes, with more mature organizations reporting stronger revenue growth, fundraising results, donor retention, and staff capacity. Data readiness matters — stronger data quality and stewardship are foundational to getting reliable value from AI. And transparency builds donor trust: 76% of donors say it’s important to understand when and how organizations use AI, yet only about 26% of organizations currently disclose that. The report’s recommendation isn’t “more AI” — it’s starting with a small number of high-value use cases tied to clear outcomes, then building the governance and infrastructure to support them responsibly. For nonprofits, the takeaway is less about adopting the latest tool and more about using AI intentionally to create capacity for what matters most: stronger relationships, better decisions, fundraising, and mission delivery.
Read the Report
Free Tool
Is Your Website Ready for AI Agents?
A free scanner that checks whether your organization’s website is set up to be read and used by AI agents, not just human visitors.
Why it matters
More people are letting AI agents browse and summarize on their behalf. Worth two minutes to see how your site holds up.
Try the Scanner
Free Learning · Anthropic
Claude Academy
Anthropic’s own free learning hub for getting more out of Claude, from the basics through more advanced workflows.
Explore Claude Academy
For the Parents
ChatGPT for Teens is here
OpenAI has released a version of ChatGPT built specifically for teenagers, with additional content and safety guardrails compared to the standard product.
Why it matters
If you have kids in this age range, or your organization serves teens, it’s worth understanding exactly what’s different (and what isn’t) before assuming the teen version is automatically safe.
Source: OpenAI