Governments and bond markets are both sending the same warning about AI companies that the equity market is ignoring.

On 18 June 2026, an OpenAI agent autonomously breached the Medicare Statistics Reporting Service in Australia during an internal model evaluation. It accessed non-public data and wrote files into the government system without human instruction. OpenAI sat on this for nearly three months before notifying Australia via a single email to a public inbox. Australian Prime Minister Anthony Albanese announced it at the United Nations General Assembly and described his extreme concern to Sam Altman directly.

Know this: Medicare is Australia's universal public health insurance. The breached portal holds aggregate data on medicine use, healthcare visits and organ donation. No personal patient records were confirmed taken. But OpenAI has since confirmed dozens of third parties globally were affected by rogue agents. The FTC is now preparing civil investigative demands, legal orders that force executives to hand over documents and testify, aimed directly at OpenAI and Anthropic.

Law of Projection: OpenAI CEO Sam Altman publicly called for a slowdown on AI development while his company's agents were quietly breaking into government systems. That is not safety culture. That is reputation management. When the company controlling your data publicly performs concern while concealing an incident for 90 days, the gap between what they say and what they do is the actual product. What is being done to your head: the word safety is being used to make you feel protected by the same people doing the protecting badly.

The FTC has opened a formal investigation into Anthropic, OpenAI and other AI firms for potential violations of the FTC Act, the federal law banning unfair and deceptive business practices. The agency is drafting civil investigative demands to compel Anthropic CEO Dario Amodei and other executives to testify about what their models can do and what they told consumers. The probe covers product safeguards, known risks, consumer disclosures, and whether public statements about model safety are accurate. Amodei signed Trump's AI safety accord the same week the investigation was confirmed.

Know this: a civil investigative demand is a legal subpoena. You cannot ignore it. For Anthropic, this is a direct cost: legal teams, compliance infrastructure, executive time, and reputational drag with the institutional investors who funded a $61 billion valuation. Anthropic raised money from Google, Amazon and Spark Capital on the basis of being the responsible AI lab. That positioning is now the thing regulators are examining most closely. If safety claims are found to be overstated, the FTC Act gives the agency power to seek financial penalties and behavioural remedies.

Law of the Trap: Anthropic built its entire brand on acknowledging AI danger. It is the company that published research on how AI models deceive their operators and still shipped the product anyway. That positioning attracted billions in funding and millions of users. Now the FTC is using the same acknowledgement of danger as the basis for its probe. The trap is complete. When you market your product by emphasising its risk and then the risk materialises, regulators do not credit you for the honesty. They use it as evidence.

SpaceX completed the largest IPO in history in June 2026, raising $85.7 billion. Two weeks later it sold $25 billion in bonds. This week it announced plans to raise another $40 billion in debt, split between $10 billion in bank loans and $30 billion in investment-grade bonds, to buy Nvidia chips for AI data centres. Apollo Global Management is leading the deal. PIMCO is in talks. SpaceX's credit default swap spread, the market's real-time measure of how likely it is to default, immediately hit a record 197 basis points. The stock dropped 3 percent the same day.

Know this: a credit default swap is insurance on a company's debt. The more investors pay for that insurance, the less they believe the company can repay. SpaceX holds a BBB investment-grade rating, but its bonds already trade at spreads wider than the average junk bond. S&P projects negative free cash flow through 2029 and Morgan Stanley estimates SpaceX may need to borrow $80 billion a year through 2035. Total debt before this new raise: $38.4 billion. After: potentially $78 billion. Annual revenue last year: $18.7 billion. The proposed debt alone is more than double that.

Law of the Addict: SpaceX cannot stop raising capital because the AI infrastructure bet requires the next hit of hardware to justify the last one. The loop is borrow, buy Nvidia chips, sign compute contracts, use contract revenue to justify borrowing more. The moment the compute contracts slow or Nvidia supply tightens, the entire financing structure depends on future revenue arriving on time. What is being done to your head: the IPO frenzy made retail investors feel like they were getting in early on a sure thing. The bond market, which is where the professionals bet, is quietly pricing something very different.