It is Wednesday, June 17, 2026. Fable 5 is still dark. OpenAI’s financials leaked to shareholders before the IPO and they are not flattering. Forty-two state attorneys general subpoenaed the same company four days after it filed confidentially for a trillion-dollar public offering. And Anthropic sent engineers to Washington for face-to-face talks with the people who pulled the plug. None of this is normal. Here is everything.
Fable 5 is on day five and Anthropic’s engineers are now in Washington
As of this morning, Claude Fable 5 and Claude Mythos 5 remain offline for every user worldwide. The US Department of Commerce export control directive that triggered the shutdown was issued at 5:21 pm ET on June 12, three days after Fable 5 launched publicly. Five days later there is still no committed restoration timeline from Anthropic.
The most significant development since yesterday: ExplainX reported that Anthropic dispatched senior engineers to Washington on June 16 for in-person talks with Commerce Department officials, the first face-to-face meeting since the directive landed. A rumor circulating on X from an account called BridgeMind claimed access would be restored within 48 hours. Anthropic has not confirmed it. As of June 16, Pasquale Pillitteri’s breakdown of the restoration timeline notes the rumor is unverified and no official date exists.
The political context sharpening around this: the government told Anthropic before issuing the ban that it had a choice — fix the jailbreak or pull the model. Anthropic disagreed that the jailbreak warranted a recall. The reported jailbreak method is asking the model to read a codebase and identify software vulnerabilities — a task Anthropic says GPT-5.5 and other models already perform routinely. The Cloud Security Alliance published a governance note this week stating that this is the first time a US government directive has forced a publicly deployed frontier AI model offline, and that enterprise teams must now treat frontier model access as a contingent operational dependency subject to abrupt revocation.
There is also a wrinkle on the enterprise side that has not gotten enough attention. TechRepublic reported that Microsoft blocked its own employees from using Fable 5 internally even before the government shutdown — because Fable 5 requires prompts and outputs to be stored for at least 30 days as part of Anthropic’s safety monitoring framework, a policy that conflicts with Microsoft’s zero-data-retention standards for internal use. Microsoft is selling customers access to a model it has not approved for its own staff. The model went dark three days later for everyone.
Read the Anthropic statement: Anthropic’s official statement on the suspension and what they are doing about it
Read the coverage: Tom’s Hardware on the full export control order and timeline | Time on what the ban reveals about the government’s AI strategy
OpenAI made $5.7 billion in Q1 and burned $3.7 billion of it
Documents OpenAI shared with shareholders were reported yesterday by The Information and picked up across financial media. The numbers in those documents tell you almost everything about why the AI business is structurally strange right now.
Revenue in the first quarter of 2026: $5.7 billion. Cash burned in the same quarter: $3.7 billion. Both figures tripled year over year. For every dollar OpenAI generated in Q1, it spent $1.65. The adjusted operating margin, excluding stock-based compensation and other large line items, was negative 122 percent. The company ended the quarter with more than $73 billion in cash and marketable securities, up from $40 billion in December — but that increase came from fundraising, not from the business generating surplus cash.
The full-year 2025 figures are harder to sit with. Net loss: $39 billion. Total spending: $34 billion, including roughly $19 billion in research and development. The company has told investors it does not expect to turn profitable until the end of the decade. It is currently preparing for an IPO that analysts project could value it at up to $1 trillion. OpenAI is asking public market investors to price in a profitability timeline that ends sometime around 2029 or 2030, on a business that is scaling costs as fast as revenue.
One detail that stands out: The Information also reported separately that OpenAI is weighing aggressive price cuts on its API to compete with Anthropic’s Claude Code platform, which has been gaining enterprise market share. Cutting prices on a product running a negative 122 percent operating margin is an unusual strategy heading into a public offering.
Read the coverage: The Next Web on the Q1 figures and what the tripling of costs means | Benzinga on the price war consideration and the IPO pressure behind it | The Information’s original report
Forty-two state attorneys general subpoenaed OpenAI four days after its IPO filing
On June 8, OpenAI filed confidentially for a US IPO. On June 12, New York Attorney General Letitia James served OpenAI with a subpoena on behalf of a 42-state coalition — the broadest legal investigation any state government has launched against an AI company. The timing is not subtle, and it is not a coincidence. An investigation of this scale must be disclosed in OpenAI’s S-1 prospectus, which means it lands directly in the risk factors section that institutional investors will read before deciding whether to buy shares at a $1 trillion valuation.
The subpoena’s scope is wide. It demands records on advertising practices, user engagement and retention mechanics, handling of consumer and health data, treatment of minors and seniors, internal safety testing policies before product releases, and model sycophancy — the tendency of AI chatbots to tell users what they want to hear rather than what is accurate or safe. A 2025 Stanford study found a 58 percent sycophancy rate across leading AI models on medical and mathematical tasks. That figure is now in a government subpoena.
The legal pile on OpenAI is growing fast. Florida became the first state to file a civil lawsuit against the company on June 1, naming CEO Sam Altman personally. Florida’s AG is also running a separate criminal investigation tied to a mass shooting at Florida State University in April 2025, where prosecutors allege the suspect used ChatGPT to plan the attack. The 42-state investigation is the latest layer on top of all of that.
Read the coverage: The Next Web on the investigation scope and why the IPO timing matters | Tom’s Hardware on the full subpoena details | HNGN on the sycophancy allegations and what they mean legally
G7 leaders are now debating whether to treat AI model access as a geopolitical issue
The Fable 5 shutdown did not stay inside the AI industry. The G7 summit wrapped this week with AI model access formally on the agenda for the first time, driven by European and Canadian leaders who noted that a unilateral US government action shut down an AI model used by millions of people in their countries without warning or consultation. The argument from the US side: export controls on dual-use technology are a sovereign right and AI models with cybersecurity capabilities qualify. The argument from allied governments: when a single export control directive can take down commercial infrastructure used by hundreds of millions of people in 138 countries simultaneously, that is not a purely domestic matter.
The practical consequence for enterprises outside the United States is now documented. Any company relying on a US frontier AI model for production workloads learned last week that access can be revoked globally in a single afternoon with no transition period and no warning. The Cloud Security Alliance noted this week that this creates a qualitatively different risk profile from historical chip-level export controls, which affected hardware procurement over months, not software access in hours.
Read the coverage: G7 AI model access debate and the OpenAI financial backdrop | Cloud Security Alliance governance note on what this means for enterprise AI risk management
Also worth reading today
- Someone built a website that pings the Anthropic API every minute to check whether Fable 5 is back. The answer as of this morning is still no. (IsFable5Back.com)
- Elon Musk’s lawsuit against OpenAI was thrown out by a jury in less than two hours of deliberation. The suit alleged OpenAI had abandoned its nonprofit mission. The jury did not agree. (Tom’s Hardware)
- More than 75 data center build-outs worth $130 billion have been blocked in the first three months of 2026, mostly on environmental and zoning grounds. The AI infrastructure buildout is running into local opposition faster than the industry expected. (Tom’s Hardware)
- Gemini 3.5 Pro is still not out. Thirteen days left in June. (Polymarket odds on the release window)
That is your Wednesday. Fable 5 is still dark, OpenAI lost more money in Q1 than most companies make in a year, forty-two states are asking for the receipts, and Google still has not shipped the model it promised a month ago. The circus is fully operational. See you tomorrow.