On Wednesday, the labor-media group More Perfect Union published audio from one of Andrew Bosworth's weekly internal meetings at Meta. On the recording, the CTO tells employees: "We fine and fire leakers all the time. We tend not to sue those people, but I'd like to..." The threat made its way to the public as a leak, from the same meeting, within days. Meta hasn't commented on the recording.

What makes this more than an irony is that Meta has already run the experiment. In early 2025, Zuckerberg complained to an all-hands that "everything I say leaks. It sucks"... and that meeting leaked. The security chief circulated a memo threatening termination for leakers... The Verge obtained the memo. The company fired roughly 20 people over leaks, and Bosworth said they were getting better at catching them. Eighteen months of enforcement later, a labor group is publishing a running series of Meta's internal audio, and the newest clip is the CTO wishing he could sue people. So enforcement went up and leaking went up with it. Whatever the anti-leak crackdown is for, stopping leaks isn't what it does.

The research on whistleblowing explains why, and it's been consistent for decades: employees go outside when raising things inside feels futile or dangerous, and each failed or punished attempt pushes them more toward an independent outlet.

Look at what Meta's employees tried first. Per IBTimes UK's reporting, when Meta put monitoring software on US work laptops in April (keystrokes, screenshots, no opt-out) more than 1,600 employees signed an internal petition. Meta says the data was tightly controlled and never used for surveillance, and it paused the program June 22 after a security review reportedly found the material sitting open across some 45,000 internal database tables. But in the leaked audio, Bosworth says the workers who pushed back "made the entire company worse." That's the sequence the research predicts: internal voice, met with hostility, becomes external voice. So the leaks aren't the disease itself... they're the symptom of a deep culture of mistrust.

Which gives you the argument to make for yourself when your own leadership wants to apply the enforcement playbook: by all means protect genuinely confidential material, but understand that punishing leakers without fixing why people leak just reroutes the traffic. At Meta it rerouted straight to a media outlet. The credible alternative is treating each leak as a diagnostic: what did this person try internally first, and what happened to them when they did? Meta's own CTO told a June meeting that morale is "probably one of the worst it's ever been" in his two decades there. So he has the diagnosis (culture of distrust) but the open question is whether anyone there wants it.

On Tuesday morning, IBM announced a bad quarter a week early, in a signed letter from the CEO, with the worst numbers spelled out in his own voice. "These conditions require our teams to execute perfectly, and this quarter we faltered," Arvind Krishna wrote to investors. "These are not excuses, but they are realities." By the close, IBM had lost 25 percent of its value, its worst single day in 115 years, beating the record it set in the crash of October 1987. Former Cisco CEO John Chambers told Business Insider the transparency will build trust, and he's probably right about the letter. But the letter is one data point in a year of IBM communications.

Walk the tape backward.

In February, IBM had its worst day since 2000, down 13 percent, after Anthropic released a tool that automates analysis of COBOL, the sixty-year-old banking language IBM still services; Krishna's public line was that AI is a growth multiplier for IBM, not a threat. On the April earnings call he told investors the first quarter was "a strong start to the year" and that "AI continues to be a tailwind for our global business," guiding to 5 percent-plus growth for 2026. On July 8, IBM launched Lightwell, a $5 billion security initiative with a who's-who of bank customers attached. Six days later: "we faltered." Every public statement IBM made this year pointed one direction, and then this letter had to turn the whole thing around in a single morning. That's the catch with owning bad news: it only lands if people weren't just told the opposite. Twelve weeks ago, Krishna was telling investors everything was on track.

We don't know whether IBM said anything to its own people, no internal note has surfaced. We do know employees are worried. On one anonymous layoff forum, a poster wrote that Red Hat "is carrying all of IBM now, if it falters turn out the lights lock the door and go home," and others were doing the math on stock compensation that vested right into the drop. "We faltered" reads as accountability to a shareholder and as an accusation to the person who worked the deal that slipped.

Here's what the next two weeks should look like if IBM's comms team is driving. First, July 22 has to be the full reset, taken once; the letter explicitly deferred full-year expectations to that call. Second, the call's most valuable disclosure costs nothing: the letter said large deals "failed to close on the timelines we expected," so say specifically how many have closed in the four weeks since. That single number converts "trust us" into something checkable by Q3, which is what candor buys. Third, retire the "tailwind" framing and tell the AI story whole: the market has now sold IBM off twice this year on the same fear (February on the COBOL tool, July on clients diverting spend to AI hardware) and IBM has answered each one separately while never addressing them as one narrative. Krishna should name the repricing threat himself, next to the Red Hat and consulting numbers that cut against it, because a story your CEO won't tell gets told by analysts at maximum damage. And fourth, before the call, not after, a Krishna-signed internal note with a manager cascade that answers the questions burning through online forums: what "we faltered" means for performance reviews, and what happens to compensation for people whose equity vested into a 25 percent hole. That's the part IBM can still fix by next Wednesday.