I've watched a lot of enterprise deals go sideways. Bad pricing, misaligned incentives, scope creep that turns a six-figure contract into a seven-figure liability. But I've rarely seen a single deal simultaneously trigger a senior leader resignation, a 200% spike in product uninstalls, and hand your biggest competitor the #1 spot on the App Store.
OpenAI managed all three in under a week.
What Happened
Here's the timeline, because the speed of this collapse is the lesson:
Late February: Anthropic's $200M Pentagon contract collapses after the company refuses to drop its red line against lethal autonomous weapons. The Trump administration blacklists Anthropic.
Friday, March 7: OpenAI announces it has signed a deal with the Pentagon for "classified AI deployments." The agreement lands with the subtlety of a flashbang.
Immediately: Backlash erupts. According to Sensor Tower data reported by the BBC, ChatGPT daily uninstall rates surge 200% above normal. Anthropic's Claude AI jumps to #1 on Apple's App Store — despite being blacklisted by the government.
Monday, March 3: Sam Altman admits on X the deal was "opportunistic and sloppy." OpenAI amends the contract to explicitly prohibit domestic surveillance of U.S. persons and bar NSA access without a separate agreement.
Today, Saturday, March 8: Caitlin Kalinowski, a senior member of OpenAI's robotics team, resigns on principle. Her stated concerns: "surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got."
The Real Problem Isn't the Pentagon Deal
Let me be direct: the problem isn't that OpenAI signed a government contract. Government AI contracts are legitimate business. The Department of Defense is the world's largest employer. AI will be part of national security infrastructure whether we like it or not.
The problem is how they did it.
Kalinowski herself drew this distinction. She said she had "deep respect for Sam and the team" and that "AI has an important role in national security." Her issue was process: policy guardrails weren't sufficiently defined before the announcement.
This is a textbook case of optimizing for deal velocity over internal alignment.
A CTO I know calls this "closing the deal before you've closed the room." You celebrate the revenue before your own people understand what you just committed to. And then you spend the next two weeks doing damage control instead of delivery.
The Enterprise Math That Should Terrify Every CEO
Let's talk numbers, because this is where it gets real.
The uninstall spike: A 200% increase in daily ChatGPT uninstalls isn't just a PR problem — it's a customer acquisition cost problem. Every user you lose costs you the marketing spend that brought them in, plus the lifetime value they would have generated. For a consumer AI product competing on network effects and usage data, this is structural damage.
The App Store inversion: Claude hitting #1 means Anthropic didn't just benefit from OpenAI's stumble — they captured intent. Users actively sought an alternative. That's not passive churn; that's motivated switching. Those users are harder to win back.
The talent cost: Kalinowski isn't a random employee. She's a senior robotics leader — exactly the kind of person OpenAI needs as it pushes into physical AI and hardware. Replacing technical leaders takes 6-12 months and costs multiples of their compensation in lost institutional knowledge and team disruption.
The contract value vs. brand damage: OpenAI hasn't disclosed the contract value, but even if it's hundreds of millions, compare that to the risk: consumer trust erosion in a market where trust is the product. When your users trust you with their queries, their data, their workflows — and then you rush into classified military deployments without clear guardrails — you're trading long-term brand equity for short-term revenue.
"Opportunistic and Sloppy" — The Five Words No Enterprise Leader Should Ever Have to Say
Sam Altman's admission is remarkable. "Opportunistic and sloppy" is exactly how you describe a deal that was closed because it was available, not because it was ready.
In enterprise sales, this pattern is common. A competitor stumbles (Anthropic loses the contract). The opportunity appears. The pressure to move fast is enormous — someone else will grab it. So you sign before the ink on your own policies is dry.
I've seen this play out at Fortune 500 companies. A government RFP opens up, the sales team moves at lightspeed, and legal/compliance/engineering are brought in after the handshake. The deal closes. Then the questions start: Can we actually deliver this within our ethical framework? Did we check with the people who have to build it?
Kalinowski's resignation answers that question for OpenAI: No. They didn't check.
The Anthropic Contrast
Here's what makes this story particularly sharp. Anthropic lost a $200M Pentagon contract because they refused to budge on their red line against lethal autonomous weapons. They got blacklisted by the government. And then their competitor swooped in, grabbed the deal, and promptly imploded.
The result? Anthropic is now #1 on the App Store. Their principled stance — which looked like a business liability two weeks ago — turned into their biggest competitive advantage.
There's a lesson here that goes beyond AI ethics: some revenue isn't worth the cost. Not because of morality (though that matters), but because of math. The reputational damage, talent attrition, and customer churn from a misaligned deal can exceed the contract value by orders of magnitude.
In conversations with security leaders over the past few years, I've heard this principle stated bluntly: "We'd rather lose a deal than lose our people." It's not idealism — it's retention economics.
What Enterprise Leaders Should Do Right Now
If you're running a company that might face this kind of decision — a lucrative contract that tests your values, your team's comfort, or your customers' trust — here's the playbook:
1. Define your red lines before the deal shows up. OpenAI had to retroactively add guardrails against domestic surveillance and NSA access. Those should have been non-negotiable before the first meeting. Your ethical framework isn't something you figure out during contract negotiation.
2. Close the room before you close the deal. Brief your technical leaders, your ethics team, your senior ICs. If Kalinowski had been part of the deliberation before the announcement, she might still be at OpenAI. Internal alignment isn't bureaucracy — it's risk management.
3. Calculate the full cost of the contract. Revenue is one number. Brand damage, talent attrition, customer churn, competitive repositioning — those are the other numbers. A $200M contract that costs you $500M in brand equity is a bad deal.
4. Speed is not a strategy for sensitive deals. Altman admitted they rushed to "get this out on Friday." In government contracting, especially anything involving classified deployments and military applications, speed signals recklessness, not capability. Deliberation signals maturity.
5. Watch what your customers do, not what your board says. A 200% uninstall spike is your customers voting with their feet. That signal is more important than any boardroom approval.
The Bigger Picture
We're in a moment where AI companies are being forced to choose: government revenue or consumer trust. Military contracts or ethical brand positioning. Speed or deliberation.
The answer isn't always obvious. But the process should be. Define your principles. Align your people. Calculate the real cost. And if a deal requires you to say "opportunistic and sloppy" five days later — that deal wasn't ready.
Kalinowski understood this. She wrote that she cared deeply about her team and the work they built together. She said it wasn't an easy call. But she made it.
The question for every enterprise leader is: would your people make the same call? And if they did — would you have given them reason to stay?
Continue Reading
[AI & Government]:
- Anthropic's Pentagon Contract Collapse: What Enterprise Leaders Need to Know — The vendor risk implications of government AI blacklists
- AI Agents in the Enterprise: What Actually Happens Next — Separating agent hype from deployment reality
- Broadcom's $100B AI Forecast — Where enterprise AI infrastructure spending is actually going
Know someone who'd find this useful?
Forward this email to a colleague who's navigating the AI landscape. They can subscribe at beri.net/#newsletter — it's free, twice a week, and I read every reply.
If you were forwarded this, click here to subscribe.
— Rajesh