AP Licenses Its Archive on 1-3 Year AI Terms. Deal Duration Is the Lever.

The Associated Press caps its AI licensing deals at one to three years. In a July 2026 McKinsey interview, CEO Daisy Veerasingham described a rights-holder strategy that most training-data commentary ignores: take 180 years of reporting, restructure it into a data product, and license it out on terms short enough that the archive is never locked into a model AP cannot renegotiate. Deal duration is the control mechanism.

Almost every governance argument about training data is written from the builder’s chair. What data can a model ingest, under what license, with what audit trail. AP is playing the other side of the table. It owns the corpus. Its question is a rights-holder’s question: how to sell access to a durable asset without surrendering it permanently.

The archive becomes a data business

AP is turning 180 years of journalism into a structured and unstructured data business sold across industries, per the McKinsey interview. That framing matters more than it first appears. A news wire has historically sold stories: a finished article, a photo, a video package delivered to a subscriber who publishes it. A data business sells something different. It sells the corpus itself, structured for machine consumption, priced by access rather than by headline.

The reframe changes what AP is protecting. When you sell an article, the transaction ends at publication. When you license a corpus for AI training, the value transfers into a model that may run for years and that you no longer control. The asset stops being a stream of daily output and becomes a standing reserve. Veerasingham’s stated design builds the licensing terms around protecting the journalists’ underlying intellectual property, not just monetizing yesterday’s news.

Consider the composition of that reserve. Eighty percent of AP content is visual, per the same interview. Photography and video are harder to synthesize convincingly and harder to source elsewhere at wire-service scale and provenance. A visual-heavy archive with a verifiable chain of authorship is exactly the kind of asset a model builder cannot easily replace. That scarcity is what gives a short-term licensing posture its leverage.

Duration as the governance lever

A perpetual or long-dated AI license hands the rights-holder’s most durable asset into a system it can no longer price, audit, or withdraw. Models trained on a corpus do not forget it when a contract lapses. So the only moment a content owner holds real leverage is before signing, and the length of the term decides how often that moment returns.

A one-to-three year ceiling does several things at once. It forces renegotiation while the asset still has scarcity value, before synthetic alternatives or competing archives erode the price. It keeps pricing anchored to a market that is repricing AI inputs almost quarterly, so the owner is never stuck at 2026 rates in 2031. It preserves the option to walk, to switch counterparties, or to change terms as the legal ground under training data keeps shifting. And it gives the rights-holder a recurring seat at the table instead of a one-time check.

This is the same insight we traced from the model side in training data as a governance lever, now inverted. There, the point was that whoever controls the training corpus controls the model’s behavior. Here, the rights-holder uses that same control to keep the corpus renegotiable. Short terms are how you refuse to let a buyer convert a rental into a permanent acquisition by default.

The mechanism is deliberately boring, and that is its strength. No litigation, no injunction, no regulatory dependency. Just a contract clock the owner sets and resets on its own schedule.

Why most content owners give this away

The default posture for a content owner approached by an AI buyer is to treat the deal as a windfall. A large check arrives for material already produced and already paid for. The temptation is to sign long, lock in the revenue, and move on. That instinct is exactly backwards for an appreciating, hard-to-replace asset.

Long terms optimize for certainty of revenue. Short terms optimize for control of the asset. A content owner who signs a five or seven year AI license in a market repricing every few months has traded away the one lever that survives the signature. The check clears once. The corpus keeps working inside the model for the full term, at a price frozen on the day of the worst available information.

The same coupling risk we described for platform citations applies to licensing structure. Once your asset is embedded in someone else’s system on their timeline, your negotiating position decays with every month you cannot revisit the terms. Duration is how you keep the coupling loose.

Do this now

If you own content that AI buyers want, audit your licensing terms this quarter against three questions.

First, what is the term length on every active or proposed AI license? Anything past three years, mark it. Long-dated deals in a fast-repricing market are where owners quietly lose leverage.

Second, is your archive structured to be sold as data, or only as finished output? AP’s move was to restructure the corpus itself into a licensable product. If your only sellable unit is the finished article or asset, you are leaving the more valuable form of the asset on the table.

Third, does your license protect the underlying rights or just monetize the output? A deal that transfers training value without protecting the creators’ intellectual property is a one-time sale dressed as a partnership.

AP’s approach comes from a single executive interview, not measured outcomes, so treat it as a stated strategy rather than proven playbook. The underlying logic holds regardless of AP’s results: for an asset that appreciates and cannot be easily replaced, the length of the deal is the governance decision. Set the clock short enough to stay in the room.

---

This analysis synthesizes How a 180-Year-Old News Institution Prepares for Its Next Reinvention (McKinsey, July 2026), an interview with AP CEO Daisy Veerasingham. Claims reflect AP’s stated strategy, not independently measured results.

Victorino Group helps content owners structure AI licensing so their IP stays renegotiable. Let’s talk.