Contract chasing is a safe first job for AI, and Britain expects AI ransomware by 2028

Docusign's connection lets Claude, ChatGPT, Gemini and Copilot read and act on the agreements in a company's Docusign account, and it has been generally available since the end of September. A connected assistant can list contracts that were sent but never signed, and track renewals and flag deadlines in supplier agreements. Before anything is sent, it has to stop and ask the person.

The costly problems with contracts are usually things that nobody looked at in time: a renewal that rolled over at a higher price because the notice window closed, or an order form left unsigned for three weeks. Those failures come from forgetting, and an assistant that gets the same question every Monday does not forget.

For example, a procurement manager at a retailer with 250 supplier contracts could start each week by asking for two lists: agreements sent more than ten days ago and still unsigned, and contracts that renew automatically in the next 90 days. The first list becomes reminder emails that the assistant drafts and she approves. The second list is where the money is. A $120,000 cleaning contract with a 5% yearly rise costs an extra $6,000 a year when its notice date passes unnoticed.

The approval step is what makes this a safe first job: a mistake shows up as a wrong draft in front of the person who decides. The assistant sees only what is stored in Docusign, so a contract signed on paper still needs someone to track its dates.

The AI answers the drive-thru speaker at about 900 US restaurants, which also keep their crews longer than restaurants without it.

A bigger bill usually means the second question got asked: a drink or a larger size. A crew member skips it when eight cars are waiting. The AI asks every time and leaves the crew free to make the food, which helps explain why people stay.

Any phone or counter order desk loses sales the same way. A parts distributor whose AI asks every caller about the matching seals would earn its gain from calls it already gets.

The agency's reasoning is that AI makes searching software for flaws cheap, so weaknesses tolerated for years get found at a scale human attackers never reached. Two flaws in PHP, the language behind most websites, sat in every version since 2012 until researchers working with AI found them in October.

For a small firm, being too small to bother with used to be a defence. When looking costs almost nothing, the customer portal that nobody has updated in years gets examined too. Defenders can use the same AI, so ask whoever runs that portal when it was last checked.

The 1,271 people affected will lose their jobs, change roles or move to one of Canva's main offices. Canva grew by selling an easy screen: templates that let anyone make a poster without a designer. A chat assistant goes one step further, because the user describes the poster and gets a draft without choosing a template.

Any app whose main value is an easy screen over a simple job, such as a form builder or a slide template library, faces the same pressure. Apps that hold a company's records keep something a typed request cannot replace.

The rule has applied since January to anyone who built, changed or used the AI system behind a harm. The injured side must still prove the company was careless; what has gone is the excuse.

That excuse is most tempting when AI agents hand work to each other. Picture one agent reading a supplier's ingredient sheet, a second writing the product page from it, and a third publishing the page. If the first agent misses an allergen, the other two do their jobs correctly with wrong information, and a customer with an allergy is harmed.

So the chain needs one named person who owns what comes out of the end, and a short record of what each agent passed to the next. That record is the proof of care that the company will need in court.