Spilling the Tea: Sora's $15M/Day Disaster, AI Washing, Meta's Secret Wearable Pilot

What the press releases don't say. What the Slack channels are buzzing about. What the enterprise AI world is actually talking about right now.

The Sora Shutdown Was Way More Embarrassing Than OpenAI Let On

OpenAI killed Sora — its much-hyped AI video generation product — and the numbers behind the shutdown are brutal. Estimated burn rate: $15 million per day. Total revenue generated before shutdown: approximately $2.1 million. That is not a product pivot. That is a case study in how demo-to-revenue gaps can survive inside a company with $20 billion in annualized revenue from other products, right up until they cannot.

The rumor making the rounds in enterprise AI circles: at least two Fortune 500 companies had signed letters of intent for Sora-based content workflows before the shutdown. Both are now quietly evaluating Runway and Kling AI as alternatives. The broader lesson: any AI product you are building a workflow around should have a revenue model that makes sense independently of its parent company's fundraising success.

The "AI Washing" Problem Is Getting Worse, And Everyone Knows It

Tech layoffs are averaging 1,115 jobs per day in 2026. Almost every announcement cites AI. A May 2026 Gartner study of 350 firms found that the companies making the deepest AI-related workforce cuts showed no improvement in financial returns. OpenAI CEO Sam Altman said it out loud in February: companies are blaming AI for workforce reductions they would have executed regardless.

The insider phrase for this is "AI washing" — and it has moved from skeptic blogs into mainstream acknowledgment. The practical problem for enterprise AI buyers: when a vendor tells you their product replaced 40% of your workflow, ask them how they define "replaced" and what their methodology is for calculating it. The gap between claimed and demonstrated productivity is the biggest unresolved issue in enterprise AI right now.

The MCP Land Grab Is Happening Faster Than Anyone Expected

Anthropic's Model Context Protocol crossed 97 million installs in March 2026. The Linux Foundation took it under open governance the same month. Every major AI provider now ships MCP-compatible tooling.

What this means that nobody is saying clearly: MCP is becoming the TCP/IP of the agentic AI layer. The companies building MCP servers for their enterprise products right now are positioning for a world where AI agents connect to business systems the same way web browsers connect to websites. The companies that do not build MCP compatibility into their products in 2026 will spend 2027 explaining to customers why their software requires custom integration work that competitors provide out of the box.

51% of CIOs Think AI Adoption Is Already Moving Too Fast

The Logicalis 2026 Global CIO Report surveyed over 1,000 CIOs worldwide. Results: 94% of organizations increased AI investment in the past year. 51% believe adoption is already moving too fast. 89% describe their approach as "learning as we go." Only 44% fully grasp the risks of the AI they have already deployed.

The rumor: several large enterprise AI vendors are privately acknowledging that their biggest customers are asking for slow-down provisions in new contracts — essentially the right to pause deployments without penalty while governance catches up. This is not being communicated publicly because it contradicts the "AI everywhere, now" narrative that drives valuations.

Meta's "Wearables for Work" Is More Advanced Than the Press Release Suggested

Meta acquired wearable startup Limitless and is reportedly testing an AI-powered pendant for enterprise use under the working name "Wearables for Work." The product captures ambient audio in meetings, surfaces action items in real time, and integrates with enterprise calendars and CRMs.

What the press release did not say: the early enterprise pilot reportedly included a Fortune 100 financial services firm. The product surfaced a compliance issue in a recorded meeting — a conversation that would not have been flagged by existing compliance monitoring tools. The pilot was paused while legal reviewed the implications. The product is back in testing with modified consent protocols.

The Biggest Tea of All: Amazon May Have Gotten Fable 5 Killed

On June 9, Anthropic launched Claude Fable 5, its most powerful publicly available model ever, built for long-horizon agentic work, with a 1 million token context window at $10/$50 per million tokens. Three days later it was gone.

On June 12, the US Commerce Department issued an export control directive ordering Anthropic to suspend access to Fable 5 and Mythos 5 for all foreign nationals. Since Anthropic cannot filter by nationality in real time, it shut both models down for every user on earth. This is the first government-forced takedown of a publicly deployed frontier AI model.

The part that has enterprise AI circles buzzing: according to Axios, the trigger was a demonstration, another company showed the Commerce Department a method for bypassing Fable 5's safety controls. The Wall Street Journal named that company as Amazon. Amazon is Anthropic's largest cloud partner AND a multi-billion-dollar investor AND a direct competitor through AWS Bedrock and its own AI model investments.

Anthropic pushed back, calling the demonstrated technique a narrow, non-universal method that essentially amounted to asking the model to read a codebase and identify software flaws, not a true jailbreak. The government acted anyway.

As of June 22, eleven days in, the model is still down. No restoration date has been confirmed, though Anthropic's international managing director told reporters in Seoul the company is confident access returns "in the coming days." And there is a second story buried in the system card itself that almost nobody covered, what Fable 5 was quietly doing to certain users before the government ever got involved. Read the full story: The Fable 5 Story Everyone Is Missing →