Perspective
Transparent inside, sealed outside: where your company strategy should live

Short answer: Your OKRs are your strategy written down: what you are betting on, where you are weak, and who owns what. Inside your company, everyone should be able to read them. Outside your company, nobody should, and no AI model should learn from them. Before you put that document in any tool, ask where it is stored, who can read it, and whether it can end up training someone else’s model.
What happened between Anthropic and Figma
For a few years, Figma and Anthropic were partners. Figma, like Adobe and Canva, worked with Anthropic to build Claude into its product. Mike Krieger, then Anthropic’s chief product officer, sat on Figma’s board.
Then the order of events got uncomfortable.
- April 14, 2026: Krieger resigned from Figma’s board. Figma disclosed it to the SEC the same day.
- April 17: Anthropic launched Claude Design, a tool for designs, prototypes, and slides. Many people saw it as a direct competitor to Figma.
- A few days later: at a private event, Figma CEO Dylan Field said Anthropic was “not consistently candid in their communications,” according to people in the room.
- May 28: Findell Capital, a Figma shareholder, asked Figma’s board for an independent investigation “to evaluate whether Anthropic benefitted from any improper use of Figma’s confidential information.”
That last point is a request for an investigation. It is not a finding. We have not seen any public result of an investigation, and nothing we have read suggests Figma customers’ files were involved. This is not a story about anyone stealing designs.
It is a story about trust. In July, The Wall Street Journal reported a wider backlash. Some founders started switching to other models after deciding a model provider might one day compete with them. Notion’s head of AI, Sarah Sachs, said: “Some companies might have been overly trusting.” The same report covered concerns about data retention. One of Anthropic’s newest models kept user data for 30 days for trust and safety review, and some companies worried their data could help train models that compete with them. An Anthropic spokeswoman said staff cannot read retained conversations and that the company has pledged not to use that data for training.
We are not taking sides on who was candid with whom. The useful lesson is simpler. The company you share sensitive work with today may compete with you tomorrow, or change its policies, or simply keep more than you expected. That is true of any vendor, including us. You should plan for it.
Why OKRs are the most sensitive document you have
A design file shows what you are building. Your OKRs show why: which market you are going after, which number the board is worried about, which team is behind, and what you decided not to do this quarter.
A good OKR set is honest. It names the gap. It says “we are at 40% and need 70%.” It shows who has not committed yet. That honesty is the whole point, and it is exactly what you would not want sitting in someone else’s training data or a log you cannot see.
Training on customer data is already a live legal question. In November 2025, a proposed class action alleged that Figma trained AI on files customers uploaded without asking. Those claims have not been proven. In May 2026 a federal judge sent the dispute to arbitration. Whatever the outcome, customers are now asking a question they rarely asked five years ago: what happens to my data after I type it in?
Transparent inside, sealed outside
Gahmbit is built on radical transparency. Every member of your company can read every objective. Missing coverage stays loud. There is no privacy toggle on objectives.
People sometimes hear that and assume we are casual about data. It is the opposite. Transparency only works if the walls are in the right place. We put them between companies, not inside them.
Your strategy gets its own database. Each organization in Gahmbit gets a dedicated database for people, cycles, OKRs, check-ins, and comments. It is not one big shared table with a company ID column. Accounts, membership, and billing live in a separate layer, and members of one company cannot open another company’s strategy database.
Optional AI does not learn from your OKRs. When AI is enabled for your organization, it runs on private open-source models, not a public chatbot. It does not train on your OKR text. By default, prompt and completion bodies are not kept on our billing store or at the inference provider for training. Billing keeps run metadata only: who ran which agent, when, and how many tokens it used.
What we do keep, we keep in your database. When Agents are enabled, coach conversations, Document Alignment Review reports, and workflow findings are saved in your company’s own database so your team can reopen them. That is company history, readable by your members, not a training corpus. Documents you upload for a review are not retained after the run.
AI is optional. The OKR loop is free and works with a zero credit balance. You never have to send your strategy to a model to use Gahmbit.
We want to be precise here, because vague privacy promises are part of the problem. We tell you what is stored, where, and who can read it. Our Trust page has the details.
Four questions to ask any tool that holds your strategy
You do not need a security team to check this. Ask any vendor, including us:
- Where is our data stored, and is it separate from other customers’? “Encrypted in the cloud” is not an answer. Ask whether your company gets its own database or shares tables with everyone else.
- Does any AI model train on what we write? Ask about their own models and the model provider’s. Get it in writing, and check whether it is the default or a setting someone has to switch off.
- What is kept after an AI run, where, and for how long? Prompts, outputs, logs, and debug stores each count. A good answer names each one.
- If your AI provider changed its terms tomorrow, what would happen to our data? If the honest answer is “we would find out with you,” weigh that.
If a vendor cannot answer these in plain language, that is your answer.
The bottom line
The Anthropic and Figma story will get settled, or forgotten, in boardrooms we are not part of. The question it raised for everyone else is not going away. Your strategy is worth protecting from the outside world and worth sharing with the people inside your company.
Keep your tools. Share your commitments, with your company and nobody else.
Create your free Gahmbit account. The OKR loop is free forever, with no credit card and no per-seat pricing.
Sources
- TechCrunch (April 16, 2026). Anthropic CPO leaves Figma’s board after reports he will offer a competing product
- Anthropic (April 17, 2026). Introducing Claude Design by Anthropic Labs
- Upstarts Media (April 23, 2026). How a board departure and product launch sparked an Anthropic and Figma feud
- Fast Company (2026). Design tech’s hot mess: Claude Design vs. Adobe, Canva, and Figma
- Findell Capital Management (May 28, 2026). Letter to the CEO and Board of Directors of Figma
- The Wall Street Journal (July 2026). A backlash against Anthropic is brewing in Silicon Valley
- Bloomberg Law (November 21, 2025). Figma trained AI on user data without consent, class action says
- Bloomberg Law (May 11, 2026). Startup founder must arbitrate AI training claims against Figma
- Related: Trust: how isolation works · How to see who hasn’t written OKRs · How to run OKRs without migrating your tools
FAQ
- Does Gahmbit train AI models on our OKRs?
- No. Optional AI, when enabled, uses private open-source models. It does not train on your OKR text, and prompt bodies are not kept on the billing store, at the inference provider for training, or for model training by default.
- Where does our strategy data live?
- Each organization gets its own dedicated database for people, cycles, OKRs, check-ins, and comments. Accounts, membership, and billing live in a separate layer that holds AI run metadata, not your OKR text.
- If Gahmbit is radically transparent, who can read our OKRs?
- Every authenticated member of your organization. Nobody outside it. Transparency is inside the company. Isolation is between companies.
- Does Gahmbit keep anything from AI runs?
- Yes, and we say so. Billing keeps run metadata, such as who ran which agent, when, and how many tokens it used. When Agents are enabled, coach transcripts, Document Alignment Review reports, and workflow findings are archived in your company's own database so your team can reopen them. That archive is company product history, not a training corpus. Uploaded documents are not retained.
- Do we need AI to use Gahmbit?
- No. The OKR loop is free and works with a zero credit balance. AI is optional and prepaid, only when you run it.