When a company blames AI, read the accounts.

In the first half of 2026, U.S. employers cited AI in 101,743 announced job cuts. That was about 23% of all cuts tracked by Challenger, Gray & Christmas. The number is large. It is also based on what employers said when they announced the cuts.

A layoff memo is written for workers, investors, customers, and the press at the same time. 'AI changed the company' sounds forward-looking. 'Revenue fell and costs rose' sounds defensive. Both statements may describe the same decision, but only one usually leads the headline.

The Budget Lab at Yale found no clear evidence, as of May 2026, that AI explained changes in the wider labor market. The International Labour Organization found real but uneven productivity gains and limited large-scale displacement so far. Company announcements and economy-wide causation are different measures.

A July 2026 study used S&P 500 annual filings to separate AI mentions from AI used inside business processes. It classified 11% of firms as deeply integrated and another 10% as using AI in production. Most large companies had not reached either stage.

The study found higher profit margins among deeply integrated firms. It did not find a significant association between AI adoption and productivity measured as revenue per employee. The authors warn that the relationships are descriptive and do not prove that AI caused the financial results.

A manager can still decide that a future system will support a smaller team. That is a forecast. When the company cuts people before the workflow, error rate, review cost, and customer result are measured, workers carry the risk of that forecast while executives keep the upside.

Coinbase announced about 700 layoffs on May 5, 2026, equal to 14% of its workforce. Its Form 8-K did not say AI alone caused the plan. The first stated aim was to manage operating expenses in response to current market conditions. The second was to optimize operations for the AI era.

The order matters. In Q1 2026, net revenue was $1.3 billion, down from $1.9 billion a year earlier. Adjusted EBITDA fell from $929.9 million to $303.3 million. A $65.6 million net profit became a $394.1 million net loss. Coinbase also said it would adjust its expense base up or down with market conditions.

There is another awkward fact. The same filing says technology and development expenses rose partly because average headcount in that area was 23% higher. It also says customer-support employee costs rose in part because the company moved some outsourced work into employee roles. The company was hiring, acquiring, insourcing, cutting, and reorganizing at once.

Coinbase depends on activity in volatile markets. Trading revenue can fall quickly when customers trade less or asset prices move against the company. AI may let the remaining staff handle more work. It does not explain why quarterly revenue and profit fell.

The filing supports a narrower conclusion. Management faced weaker financial results and chose a smaller operating model that it believes AI can support. That is not the same claim as 'AI made 700 jobs obsolete.' The public evidence does not identify 700 automated roles or measure the output of a system against those jobs.

This distinction protects workers from a false personal verdict. A person can lose a job because management wants a different cost structure. The event does not prove that software can perform that person's full work.

Block cut more than 4,000 people in February 2026 and planned to move from more than 10,000 employees to fewer than 6,000. CEO Jack Dorsey directly argued that intelligence tools let a much smaller team do more.

The financial results were not collapsing. Q4 2025 gross profit grew 24% year over year. Cash App grew 33%. Square grew 7%. Block reported $485 million in operating income. The same shareholder letter called 2025 a strong year.

Calling this a smoke screen for bad quarterly performance would be inaccurate. The better criticism is that the AI story made a long-running organizational choice sound technologically inevitable.

In November 2023, Block announced an absolute cap of 12,000 employees until business growth outpaced company growth. In March 2025, it cut 931 roles for strategy, performance, and fewer management layers. Dorsey wrote then that those cuts were not about a financial target or replacing people with AI.

By the end of 2025, Block reported an employee population of about 10,200. Two months later, the company said AI justified moving below 6,000. The technology may have increased management's confidence. It did not start the preference for a smaller, flatter company.

After the cut, Block reported 2.5 times more production code changes per engineer and lower incident rates. Those are useful operating measures. They are also company-selected measures from a short period after a huge reorganization. They do not tell us what products were cancelled, which maintenance work moved, or how much review the remaining staff absorbed.

When a company links layoffs to AI, read three records. Start with the layoff filing or memo. It tells you what management wants the public to believe. Then read the latest results. They show revenue, profit, costs, and the segments under pressure. Finally, read the prior annual report or reorganization memo. It shows whether the company was already cutting, freezing hiring, or removing layers.

For each claim, write what the evidence can and cannot prove.

Workers negotiate less when they believe a machine has already won. Investors reward a manager who promises more output with fewer salaries. Executives get a story that points away from overhiring, weak demand, failed bets, or years of slow decisions.

AI is changing work. The ILO's 2026 review found real productivity gains in some settings. It also found that gains were uneven, often unverified, and had not produced large-scale job displacement. That is a less dramatic finding than the layoff memo, and a more useful one.

Do not answer fearmongering with denial. Ask for the unit of work, the measured output, the time period, the cost of review, and the business results around the cut. If those facts are missing, the company has announced a theory, not proved it.