Acts of AI: Price the Disruption Before Your Lawyers Do
Buyout loans have started treating AI like a war or a pandemic. If AI risk is now in the credit agreement, it belongs in your diligence first. Here’s a six-step review a deal team can run in a week.

On this page 6 sections
War, pandemic, energy shock… and AI
A loan agreement now lists AI next to war and pandemics.
These lists live in force majeure clauses, usually the dullest page in the document: the catastrophes nobody can be blamed for. War, terrorism, pandemics, an energy shock. In late October, AI joined them.
Bloomberg reported that H.I.G. Capital got the wording into the €470 million loan behind its majority stake in Terras Group, a German infrastructure engineering firm. If Terras misses an interest payment and says AI disruption is to blame, lenders’ remedies pause for as long as it says the impact lasts.
The interest is still owed. The lenders just have to wait.
Not every lender agreed. On Bain Capital’s €4.7 billion buyout of Everllence, the former MAN Energy Solutions, lenders had similar wording taken out. The language has appeared in at least two leveraged buyout financings since September.
Lawyers are watching closely. Jeremy Duffy, who chairs European leveraged finance at Cahill Gordon & Reindel, described it this way:
“A fast-moving piece of the leveraged finance documentation pie.”
Whatever you think of the clause, notice what it admits. AI disruption is now treated as a real risk to whether a company can pay its debts, serious enough to negotiate over. Most investment committee papers still give it a paragraph.
Why it showed up now
The timing isn’t random. Software buyouts from 2021, bought at peak prices with cheap debt, are under strain.
On 4 November, Jen Kha of Andreessen Horowitz noted that many of those deals now pay interest by adding it to the loan instead of in cash. For those borrowers, loan-to-value went from 44% to 75% in a year.
- ~$60BAverage due two years out, 2023–25
- ~$140BDue in 2028
- ~$160BDue in 2029
Each brick is about $20B. Source: Jen Kha, “Is Private Equity Cooked?”, a16z, 4 November 2026
Everyone in the industry knows the cautionary tale. Thoma Bravo took Medallia private in 2021 for about $6.4 billion. This year it handed the company to lenders led by Blackstone, wiping out about $5.1 billion of equity.
Medallia’s trouble came from rate rises and an over-optimistic growth plan, not AI. But it shows how fast a software deal turns when growth slows and the debt doesn’t.
Now add AI. Many software businesses charge per user. If a customer’s team can do the same work with fewer people and a few agents, it buys fewer seats.
Investors have sold software stocks on that fear for most of this year. Put the same fear inside a heavily indebted buyout, and sponsors start asking lenders for an AI escape hatch.
The question nobody answers in the IC meeting
Picture a Thursday evening in a deal team’s war room. The people are made up; the meeting isn’t.
An associate, call her Anika, has spent three weeks on a vertical software company: churn, cohorts, pricing, the management presentation. The model is clean.
Then a partner asks the question that’s been hanging over every software deal this year: “What does AI do to this business?”
The honest answer, in most rooms, is a shrug wrapped in a slide. A paragraph saying the company is “investing in AI”, a mention of a chatbot on the roadmap, a reassuring line about high switching costs. Nobody has priced it.
That gap is what the new clauses fill. Risk nobody measures before signing gets negotiated afterwards, in the credit agreement, by people who aren’t looking at the product.
It belongs in the diligence. Here is how to put it there.
If AI risk is serious enough for the credit agreement, it’s serious enough for the diligence.
The AI exposure review
Six steps, run alongside the usual commercial and technical diligence. Most of it uses data you’re already collecting. Done properly, it takes a few days of an associate’s time and a handful of calls.
Break revenue into the work it pays for. For each product line, list the jobs customers use it for. Mark each job as one an AI agent could do today, within two years, or not soon. Price the exposed share, not the product.
Find the seat exposure. How much revenue is priced per user? How have seats per customer moved over eight quarters? A slow fall in seats with flat logo churn is the early warning.
Ask customers the uncomfortable question. “If your team used AI agents for this work, would you need as many licences in two years?” Ask what they’ve already tried. Their pilots are your forecast.
Test management’s AI plan. Separate what has shipped from what is on slides. Ask for usage data on the AI features, the price for them, and who owns the plan. A roadmap is not a moat.
Put a number on it. Build a downside case where exposed seats shrink by a range you can defend from steps 1–3. Check it against interest cover and covenant headroom, not just the exit multiple.
Read the credit documents for AI. If there is AI language, ask who decides when the company is “impacted”, when the pause ends, and whether EBITDA can be adjusted for it. Then decide whether your downside case needs it at all.
A practical framework. It complements legal and financial advice; it doesn’t replace them.
What exposed looks like
After a few of these reviews, patterns repeat. These are the signals we’d weigh most heavily.
Signs of resilience
- Priced on outcomes, transactions or assets, not users
- Holds data customers can’t easily get elsewhere
- Sits in a regulated workflow where an audit trail matters
- Already ships AI features that customers pay for
- Seats per customer stable or rising
Signs of exposure
- Most revenue priced per seat for routine knowledge work
- Value is mainly the interface, not the data behind it
- Customers already piloting agents for the same tasks
- AI plan exists mostly as a roadmap slide
- Seats per customer quietly falling
No single signal decides it. Two or three signs of exposure, together with a high level of debt, deserve a priced downside case.
Exposure isn’t only bad news. Some of the best deals this cycle will be businesses where AI makes the product more valuable, because it runs on data only that company has. The review tells you which side of the line you’re buying.
- Which three tasks do your customers use the product for most, and could an agent do any of them today?
- How have seats per customer moved over the last eight quarters?
- Which AI features are live, how many customers use them, and do they pay for them?
- What would you change about pricing if seat counts fell by a fifth?
- Who owns AI in the leadership team, and what have they shipped this year?
Back to Thursday evening
Run the review, and Anika’s answer changes. No shrug this time.
About a third of revenue comes from work agents can already do. Seats per customer have dipped two quarters running. Two of the five customers we called are piloting agents for it. Here is the downside case, and what it does to interest cover.
That answer might kill the deal, lower the price, or show the business is better placed than its peers. Any of those beats finding out later, in a clause, after the money has moved.
Lenders have started writing AI into the documents. Deal teams should get there first, with evidence.
Key takeaways
- Buyout loans now include clauses that treat AI disruption like a pandemic or a war; lenders struck one from a €4.7B deal.
- The 2021 software vintage is under strain: PIK interest, rising loan-to-value and a large maturity wall in 2028–29.
- Seat-priced software doing routine knowledge work is most exposed to AI.
- Run an AI exposure review in six steps: work, seats, customers, management plan, downside case, credit documents.
- Price the risk before signing, so it isn’t negotiated afterwards in the loan documents.
Case studies & examples
References & further reading
- Private Equity Firms Add AI Disruption Clauses to Buyout Loan AgreementsBloomberg, 27 October 2026
- Private equity seeks ‘act of AI’ escape clause in €470M buyout loanDealroom, 27 October 2026
- Private Equity Wants “AI Disruption” to Count as an Act of GodThe State of AI, 27 October 2026
- Is Private Equity Cooked?Jen Kha, a16z, 4 November 2026
- Thoma Bravo hands Medallia to lenders in one of biggest private equity restructurings everWith Intelligence, 18 June 2026
- Blackstone-led consortium takes control of MedalliaPrivate Equity Wire, 18 June 2026
- SaaS-Pocalypse explained: why AI is crushing software stocksDunham, February 2026, updated October 2026
Links checked on 5 November 2026. External sources are listed for further reading; Delphilens is not affiliated with them.