mit's finding, that 95 percent of enterprise ai pilots produce zero measurable p&l impact, gets read as a verdict on the technology. it is a verdict on the setup: a pilot with no owner, no number, and no end date cannot fail, which is exactly why it clears approval so easily.
the cost is not the license fee. it is two quarters of your best operator's attention, and the credibility you burn internally when the thing quietly stops being mentioned. s&p global found 42 percent of companies abandoned most of their ai projects in 2025, and the survivors were not the fanciest tools. they were the ones wired into a real workflow with a name attached.
the advice going around is to start small and learn, which sounds like prudence and functions as permission to not commit. do the version that holds: pick one workflow, name the number it should move and by how much, name the person who owns that number, and set the date you either kill it or scale it. if you cannot fill in all four, you do not have a pilot. you have a subscription.
a researcher found 3.7 million chat transcripts, call recordings and phone transcriptions in three unsecured databases tied to sears home services, dating 2024 to 2026, with customer names, physical addresses, emails and phone numbers inside.
the condition is ordinary, and it is probably yours. you put a bot in front of customers, the bot writes down every conversation, and the writing lives on infrastructure you have never seen and cannot audit. the cost does not land on the vendor. the name on the breach notice is yours, and the customer reading it has never heard of the company that actually left the door open.
the advice going around is to ask for the vendor's soc 2, which tells you they have a process, not where your transcripts sit or how long they are kept. do the version that holds: this week, send two questions in writing to every ai vendor that touches a customer conversation. where are transcripts and recordings stored, and what is the retention period. then set retention to the shortest window your business can work with and get it into the contract at renewal.
palo alto's unit 42 has now observed indirect prompt injection in the wild: text hidden on a page, invisible to you, read by an ai browser agent as if it came from you, while that agent is signed into your accounts. owasp's position is that this is not a bug with a patch date. it is how the architecture reads input, because the page and your instructions arrive in the same stream and nothing in the model separates them.
the cost is whatever that browser profile can reach. banking, payroll, the admin panel, the inbox you reset passwords from. and the damage looks like an action you took, from your session, with logs that agree.
the advice going around is to be careful which sites you let it visit, which assumes you know in advance every link it will follow three steps into a task. do the version that holds: give the agent its own browser profile with nothing valuable logged into it, and keep banking, payroll, admin and email in a profile it never touches. then write down the short list of actions that always require a human click. money out, credentials, anything you cannot undo.
merriam-webster made "slop" its 2025 word of the year, which is a cultural verdict and costs you nothing. the commercial verdict came from google's 2026 core updates, which went after scaled content abuse and took large chunks of organic traffic off sites publishing ai pages at volume, in some cases inside two weeks.
the cost is not the writing subscription. it is a year of budget, headcount and calendar spent producing pages no person was going to read and no machine was going to rank.
the fix being sold is a better prompt and a humanizer tool, which is the same factory with a fresh coat of paint on the sign. do the version that holds: pull your last twelve months of published pages, sort them by organic traffic and by pipeline sourced, and count how many produced neither. then stop paying by the word and fund the four or five pieces only your company could have written, the ones with your numbers, your customers and your mistakes in them.
"vibe coding cleanup specialist" is a real job title now. 404 media found coders on fiverr carrying 15 to 20 regular clients apiece, fixing apps their customers built with ai and could not ship. the customers are exactly who you would guess: sales managers, product people, and small business owners who got a working demo in an afternoon and a mess they could not touch by month two. nbc news found the same market in every other trade, illustrators repairing ai logos with garbled text in them, writers who say half their gigs are now rewriting machine copy nobody would put their name on.
the joke writes itself. the industry sold the end of paying for skilled work, and its first stable job market is skilled people billing to redo it. the part that lands on your desk is the sequencing: the draft was nearly free, so the work looked done, so it shipped or stalled, and the bill for the last mile arrived later, at a rush rate, from a specialist.
the fake move is buying a better generator, which produces a more convincing version of the same unfinished work. do the version that holds: price the review into the job before you start, name the person who signs off before anything ships, and split your outputs into two lists, what can go out unreviewed and what never does. anything with your name, a price, or a promise on it sits on the second list.