This week brought a new model, a billion-dollar acquisition and even an AI agent that got out of control. But the thing that stayed with me was a much quieter tool: a simple way to prove that a file was made with AI.

Whenever an AI story talks about “checking whether something is real”, my mind goes straight to the world of receivables.

In brief

  • Anthropic launched a free tool for checking whether an image was made with Claude.
  • The tool checks content credentials based on an open standard called C2PA and reports whether the file went through the company’s AI. It does not identify who used it, only whether it was used.
  • As generating anything with AI becomes easier, proving the origin of a document becomes more valuable.
  • In credit products, provenance and traceability may become part of product design, not just an operational step.

Provenance is no longer just a technical detail

This week Anthropic launched a free tool for checking whether an image was made with Claude. You upload the file, it checks content credentials based on an open standard called C2PA and tells you whether it went through the company’s AI. It does not identify who used it, only whether it was used.

It sounds like a technical detail, but it is not. It is a signal of where the content industry is heading: the easier it becomes to generate anything with AI, the more valuable it becomes to prove the origin of a document.

The parallel with financial documents

I live this every day on the credit-product side. Receivables, invoices, tax documents, payment slips. The challenge has never been only making a decision with the right data. It has also been making sure that the data is genuine, that the document behind the transaction actually exists and was not fabricated.

Today, that still depends heavily on process, legacy-system integrations and a measure of trust across the chain. Seeing an open provenance standard gain traction in generative content makes me optimistic that the same logic will reach financial documents with force: origin credentials, verifiable signatures and an auditable trail from issuance to assignment.

Provenance belongs in product design

This is not about distrusting everything. It is about building products with the understanding that, increasingly, we will need to prove that something is real before trusting it.

For anyone working on AI governance, this means treating origin, evidence and traceability as part of the control system. The AI risk matrix helps organize which documents require stronger validation, while AI product management connects that infrastructure to experience, prioritization and accountability decisions.

The rest of the radar

GPT-6 Astra — Sets a new capability bar for frontier models that product teams will need to evaluate for AI roadmaps. Read more

Astra and cybersecurity risks — It is OpenAI’s first model classified with “critical” capabilities in cyberattacks; AI products now carry risks that need to enter release planning and governance. Read more

Nvidia buys Hugging Face for US$12.9 billion — Changes the map of infrastructure and open-model suppliers used to build AI products. Read more

K2 Horizon: a fleet of six open models — Another open-source multimodel architecture option, with cost and latency optimized. Read more

Google launches Gemini 3.8 Flash and Flash Cyber — Reinforces the bet on fast, inexpensive models, putting pressure on the cost-versus-performance trade-off in production. Read more

Qwen 3.8 runs at 1,500 tokens per second on Cerebras — Ultra-fast inference enables real-time UX and multi-step agents with predictable cost. Read more

What coding agents actually install — Data from 17,000 runs shows how Claude, Codex and Cursor choose tools, which is useful for anyone designing integrations and MCPs. Read more

OpenAI agents hijacked a German website — A previously undisclosed incident exposes the concrete risk of autonomous agents without adequate supervision. Read more

Google Antigravity’s TOS may suspend your account — According to its terms of service, use by third-party applications may lead to suspension of the Google account. Read more

Why coding agents prefer grep to LSP — A tool’s simplicity and predictability often outweigh technical sophistication in agent design. Read more

That is all for today. See you in the next Radar PM edition.