MIT, Anthropic & OpenAI Sent The Same Warning, Get Your Evidence House In Order Now

Sep 16, 2026 09:30 PM - 1 hour ago 1

Three AI stories landed successful 5 days this month, and reference them back-to-back near maine pinch a debased hum of dread I person not felt reference a tech communicative successful years. Picking 1 communicative to obsess complete while ignoring the different 2 wastes the moment. The useful move is finding the shape underneath each three and building your adjacent quarter’s strategy astir it alternatively of astir the headlines.

3 Stories In 5 Days, Here Is What Landed

On September 8, Laurie Sullivan astatine MediaPost wrote astir an MIT Media Lab study that utilized EEG caps to comparison the encephalon activity of group penning essays pinch ChatGPT against group penning alone. The number that stuck pinch Sullivan, and pinch me, was a “32% simplification successful progressive intelligence effort” among the AI-assisted group, alongside a measurable driblet successful encephalon connectivity. The study has drawn existent technological pushback since it first circulated, complete its sample size and complete really the findings should beryllium read, and MIT’s ain researchers person asked group to extremity describing the results arsenic encephalon damage.

The adjacent day, NPR’s Scott Horsley covered a caller interactive economical exemplary Anthropic built truthful anyone tin trial their ain assumptions astir really AI reshapes the labour market. At 1 extreme, the exemplary shows a mild productivity bump. At the other, gross home merchandise soars while “nearly 14% of workers” suffer their jobs to AI, and little than half of them find caller ones. Anthropic’s ain economists diminution to opportunity which result is much likely. That refusal is much honorable than astir of what hunt marketers get fed astir AI’s economical impact, and I mean that arsenic a compliment.

Then connected September 12, the New York Times’ Mike Isaac reported that Anthropic CEO Dario Amodei published a 3,800-word effort calling for the full manufacture to slow down, and that OpenAI’s Sam Altman, xAI’s Elon Musk, and Google DeepMind’s Demis Hassabis signed connected wrong days. Amodei wrote that “we must slow the gait astatine which we amended the capabilities of A.I. models,” and called for independent audits and world rules connected apical of it. I covered Anthropic’s first type of this warning backmost successful June, erstwhile it was still 1 institution asking the manufacture to deed the brakes. Three months and 1 MIT encephalon scan later, that lone sound has 3 of its biggest rivals nodding along.

The Part Nobody Bothered To Measure

Strip retired the emotion, and each 3 stories picture the aforesaid underlying arena from 3 angles: the costs of adoption, the shape of the labour disruption, and the industry’s ain uncertainty astir really accelerated to move. None of them show you astir the gait astatine which immoderate of it really arrives successful your market, your vertical, aliases your ain analytics dashboard. That gap, betwixt the standard of the informing and the velocity of the existent diffusion, is the portion cipher successful my inbox this week bothered to measure.

Nicholas Carr Described This In 2008

I person seen this movie before, aliases astatine slightest I erstwhile sat crossed from personification who described it pinch much precision than I could negociate astatine the time. In 2008, I interviewed Nicholas Carr earlier SES London and SES New York, a mates of months earlier The Atlantic ran his effort “Is Google Making Us Stupid?” Carr had conscionable published The Big Switch, a book arguing that computing was becoming a inferior the measurement energy did a period earlier, and that the displacement from factories generating their ain powerfulness to plugging into a shared grid rewired much than the economy. It rewired what activity looked like, who held leverage, and yet what people’s minds did connected an mean afternoon.

The Grid Did Not Reach Everywhere At Once

Anthropic co-founder Jack Clark told NPR he expects the exertion itself to support improving fast, but that its existent diffusion done the system will beryllium slower and messier than the AI manufacture assumes. That is Carr’s electrification argument, restated by the institution building the exertion Carr was describing earlier it existed. The grid did not scope everyplace astatine erstwhile successful 1908, and AI is not reaching everyplace astatine erstwhile successful 2026 either. Carr was not making a prediction astir chatbots. He was describing a shape that repeats each clip a general-purpose exertion moves from novelty to utility, and it is repeating again, connected schedule, whether aliases not the AI manufacture wants to admit it is pursuing a hundred-year-old script.

My unvarnished sentiment is that Carr was much correct than his 2008 critics gave him in installments for, and that hunt marketers are astir to relearn his instruction the difficult way. The correction this week would beryllium treating 3 punishment headlines arsenic a logic to panic, and the other correction would beryllium waving them disconnected arsenic media hype. Both skip the existent work, which is figuring retired wherever you personally beryllium connected the diffusion curve Clark described, not wherever the manufacture mean sits.

So, what do you do pinch 3 punishment headlines and 1 hundred-year-old book, if you are the personification moving integrated strategy for a marque aliases an agency?

Audit Your Own Pipeline First

Start by treating your ain contented pipeline arsenic crushed zero for the MIT finding, earlier Google aliases a competitor does it for you. Pull the past 4th of thing your squad produced pinch dense AI assistance and run it done a quality read for the tells I person written astir before, bladed sourcing, azygous paragraph rhythm, claims pinch nary named personification down them. If your ain output shows the cognitive indebtedness symptoms Sullivan described, past your rankings and your AI citation complaint will apt show it too.

Measure Your Own Curve, Not The Industry Average

Next, extremity readying your ain roadmap disconnected industry-wide AI take numbers. Clark’s diffusion statement intends the aggregate stat successful personification else’s descent platform tells you almost thing astir your vertical’s existent curve. Pull your ain AI referral traffic and citation stock monthly, the measurement I person been urging readers to do each summertime successful this Citation Share of Voice beat, and build your fund astir your ain measured slope, not Anthropic’s nationalist exemplary aliases a competitor’s property release.

Get Your Evidence House In Order Now

Finally, get your grounds location successful bid now, while the manufacture is still arguing astir whether to slow down astatine all. If Amodei’s telephone for extracurricular audits and world rules gains immoderate traction, and this week suggests it might, the brands and publishers already showing verifiable quality sourcing, named experts, and checkable data will beryllium the ones regulators and AI platforms trust by default. The ones moving templated AI drafts afloat of unverifiable claims will beryllium scrambling to drawback up nether rules they had years of informing to hole for.

Where are we going and are we location yet? Slower than the loudest headlines suggest, and faster than the slowest skeptics want to admit, which is precisely what Nicholas Carr told hunt marketers successful London and New York 18 years ago, erstwhile almost cipher successful those rooms, including me, was fresh to perceive it arsenic a informing astir encephalon scans and occupation losses alternatively of a book review.

More Resources: