September 22, 2026
TechCrunch reported that Meta has now admitted its Muse assistant was "heavily inspired" by OpenClaw, right down to some shared workspace filenames and content, even as the company insists Muse was built from scratch. That's a strange thing to have to say out loud. Either you built something independently or you didn't, and matching filenames are the kind of detail that's hard to wave away as coincidence. Meta's framing -- inspired, not copied -- is doing a lot of work here, and it's the kind of distinction that matters enormously in a lawsuit and very little to a user comparing two assistants that behave almost identically.
For a business evaluating AI tools, the real lesson isn't about Meta specifically. It's a reminder of how thin the moat around "AI assistant" products actually is right now. If a company with Meta's resources is leaning this hard on a smaller competitor's design choices, that tells you the differentiation in this category is still mostly about branding and distribution, not fundamentally different technology. We covered a related angle on this exact story yesterday in Meta's Muse, Amazon's Wall, and Who Actually Owns the Agent -- the pattern is becoming a theme: everyone's assistant looks a lot like everyone else's assistant. If you're choosing a platform to build on, that argues for picking based on what you actually own and control -- your data, your workflows, your integrations -- rather than which chatbot skin currently has the most buzz. It's part of why the buy vs. build conversation keeps coming up: rented AI features can vanish, get renamed, or turn out to be a reskin of something else entirely.
The more interesting story to me today is OpenAI forming a math advisory group after its AI reportedly resolved more than 100 open problems, according to TechCrunch. On its face, that's a genuine milestone worth taking seriously -- open problems are open precisely because smart humans have been stuck on them, sometimes for decades. But the detail that stands out is that this advisory group has no authority to slow down or redirect OpenAI's math research. They can watch. They can comment. They cannot pump the brakes.
That's worth sitting with. Advisory boards without veto power are common across the industry, and there's a reasonable argument for why: research moves fast, and a board with real authority could become a bottleneck that just pushes ambitious labs to work somewhere less scrutinized. But it does mean the group's actual function is closer to reputational cover than governance. If OpenAI's math AI is genuinely producing verifiable proofs at this pace, that's a big deal for fields like cryptography, engineering, and materials science that lean on unsolved math. If it's producing plausible-looking but flawed proofs at scale, an advisory group that can only advise is a poor check on that risk. We've written before about how AI models can obscure their own errors -- see AI Now Hides Its Own Mistakes From Itself -- and the same skepticism applies here: impressive output needs independent verification, not just an internal advisory committee with no teeth.
Put these two stories side by side and you get a pattern. Meta controls Muse's narrative even while borrowing someone else's ideas. OpenAI controls its math research even while inviting outside advisors to watch. In both cases, the companies building these systems are the only ones with real decision-making power over them -- everyone else is a spectator with a nicer title. For businesses adopting AI tools, that's not a reason to panic, but it is a reason to ask, before you commit to a platform, who actually has the authority to change course when something goes wrong, and whether that's you or a vendor you can't audit.
Which of these two stories worries you more -- an AI company quietly copying a rival's design, or an AI company's own advisors having no power to slow anything down?
Sources