Meta is preparing to put artificial intelligence in charge of small-business tasks, even as fresh scrutiny lands on how the company measures its AI's abilities.
According to a report carried by MSN (via Bing News), Meta has announced an AI Business Agent that, starting August 1, is meant to handle customer questions, recommend products, and book appointments. The same report frames a pointed doubt: if Meta's AI can't reliably manage a basic Instagram account, how will it be trusted to run a business?
That skepticism is compounded by questions about Meta's benchmark claims. According to WION (via Google News), Meta once "faked" an AI benchmark score, and the outlet raises whether the company is doing the same with a model referred to as Muse Spark 1.1. WION's headline captures the stakes with the phrase "From 2nd to 32nd" — pointing to how far a ranking can fall once numbers are re-examined.
Benchmarks are the standardized tests the AI industry uses to rank how capable a model is. When those scores are inflated or gamed, buyers and business owners can be misled about what a tool can actually do in the real world — a gap that matters far more when the software is answering customers and booking appointments rather than just topping a leaderboard.
The available sources are brief and raise concerns rather than confirm technical details, so specifics about how the Business Agent performs, or the exact nature of any benchmark dispute, remain thin. Neither item, as presented, includes a direct response from Meta beyond its own product announcement.
Why it matters: as Meta pushes AI toward hands-on business roles, questions about whether its performance claims can be trusted go straight to whether that automation is safe to rely on.