Anthropic caused a bit of a firestorm this week.
The company behind Claude announced it is now adding invisible, machine-readable watermarks to AI-generated text, along with provenance metadata for generated files. You can read the details here.
My social media feeds filled up with people celebrating the prospect of catching AI users, alongside people horrified that they might be caught.
That reaction interests me more than the watermark.
If you use AI to help with your work, why are you worried that somebody might find out?
What Anthropic is actually doing
On August 2, new transparency provisions under the EU AI Act became applicable. Among other things, providers of generative AI systems are required to make AI-generated or manipulated content detectable in machine-readable form.
Anthropic has signed the EU’s Code of Practice and says new Claude models will include machine-readable marking from launch. Generated text can contain an imperceptible watermark, while supported files can contain digitally signed provenance information. Anthropic intends those markings to apply globally wherever supported models are used, not just inside Europe.
A Claude watermark doesn’t necessarily mean “Claude wrote this.”
Anthropic says marked content may have been proofread, translated, summarized, converted, or otherwise processed by Claude. Heavily editing something can remove the mark, while unmarked content might still have been generated by AI.
It is not a magical AI lie detector.
Use the tools
The loudest voices tend to occupy the extremes.
One group treats AI use as lazy, dishonest, or evidence that the work must be inferior. The other insists AI will replace practically everything, probably by next Tuesday, and anyone who has not automated their life is doomed.
Neither group is much help to someone trying to do good work.
I use AI. I’m a certified AI Governance Professional (AIGP), so hiding AI usage would be a particularly weird move.
I use modern tools because they help me think, learn, analyze, and solve problems I would struggle to tackle as effectively on my own. I don’t find that embarrassing.
Having AI skills matters, both in your work and personal lives. That does not mean using ChatGPT for everything. Security, privacy, intellectual property, professional obligations, and your organization’s policies still apply. Ignoring those rules is a bad gamble.
Refusing to learn these tools because someone might judge you is also a bad gamble.
Use AI to improve your judgment
Much AI advice focuses on completing eight hours of work in 47 minutes.
Fine. Then what?
Automating routine tasks can be useful. But proving that a collection of tasks can run without you means that you haven’t been adding a ton of value. Probably not the right look for you.
I want AI to challenge an argument, find weak spots in an idea, expose assumptions I haven’t questioned, and connect information I couldn’t realistically hold in my head at once. It can explore alternatives, handle tedious analysis, and help me communicate more clearly.
The useful test is whether AI improved my judgment and the work, or merely replaced effort I should have contributed.
Your name is still on it
Sometimes a terrible document starts circulating. The numbers don’t make sense, the recommendations are shallow, and important context is missing.
Then somebody says, “Well, that’s what AI gave me.”
Don’t ever be that guy.
If you prepared a terrible financial forecast, you would not blame Excel. Excel was a tool you used. You still prepared the forecast.
AI works the same way. Check the facts. Apply judgment. Fix the writing. Stand behind what you send.
Develop your own AI strategy
Organizations are developing AI strategies. Individuals need a practical version of their own, not a 73-page document filled with diagrams and consulting vocabulary.
Decide:
- What will I use AI for, and where will I keep it out?
- What information will I never give it?
- When must I verify its output or rely on my own expertise?
- Which employer, client, or professional rules apply?
- Could I explain exactly how AI contributed to this work without squirming?
That last question is the useful one.
If disclosure would embarrass you, examine how you are using AI. Design your use so you can say what the tool did, what you contributed, and why you stand behind the result.
Ask better questions
AI is already built into search engines, operating systems, spreadsheets, cameras, and communication tools. The boundary between AI-generated and human-generated work will keep getting fuzzier.
Anthropic’s policy shows why. A mark might mean Claude wrote something, translated it, proofread it, or reformatted it.
So judge the work:
- Is it good?
- Is it accurate?
- Did the person exercise judgment?
- Will they take responsibility for it?
Those standards worked before generative AI. They still work now.
I use AI, and I’m comfortable telling you how.