Enterprise buyers won’t put your product near their customer data without a SOC 2 report. Twenty-one state privacy laws now hold them accountable for the vendors they share data with, so their diligence lands on you.

Sprinto gets you audit ready in 14 days, across three working sessions. AI agents connect your stack, collect the evidence auditors ask for, and close the gaps as they appear. You approve, they execute.

Sprinto also answers the security questionnaires your buyers send, using the same evidence base, so security reviews stop holding up deals.

No compliance hire. No consultant. Your auditor signs off.

Learn more and get $1,000 off.

Dr. Tristan Buckmaster, professor at NYU’s Courant Institute of Mathematics, went on CNBC’s Squawk Box this morning to discuss how OpenAI may have used some of his research, fed it into ChatGPT, and set 100+ agents to beat him and a colleague to solving the Navier-Stokes problem and submitting their proof.  Dr. Buckmaster shared how he and a colleague, Levent Alpoge (who was using Anthropic’s Claude), were coming close to solving the problem.  When word reached OpenAI that a researcher using Claude was close to solving the problem, they approached Buckmaster with bribes, threats, and ultimately (he says) used his work to beat both researchers to the solution to the Millenium Prize problem. 

The NYTimes published Dr. Buckmaster’s account of the incident on September 10^th.  This morning, during his interview, Buckmaster told CNBC’s Becky Quick and Andrew Ross Sorkin that he worries most not about himself, but about young mathematicians coming onto the scene at a time in which companies like OpenAI and Anthropic are releasing models that are not only “eating software,” but indeed, all knowledge work.  Buckmaster further shared that when he resisted OpenAI’s offers to abandon his Anthropic-using colleague and work with them, an OpenAI executive said to him, privately: “Don’t you care about your career?”   OpenAI publicly stated that it’s categorically impossible for its AI system to have found and incorporated Buckmaster’s work into their proof.

722 manuscripts and a harder word than "proof."

And it seems that OpenAI is now doubling down on that claim.  The Verge reports that OpenAI has released another batch of mathematical work, 722 machine-generated manuscripts addressing long-standing problems, and that mathematicians are unsettled by it.  On the one hand, it’s good, since problems that resisted human effort for decades may be yielding. The unsettled part is subtler: a proof checked by people has an author who can explain why it works, and a proof that arrives without one raises fresh questions about who gets credit and who understands the result. Expect the field to spend the next year debating verification norms as much as the mathematics itself.  But it doesn’t stop there: as frontier labs become richer and more powerful, their agents are ingesting the “alpha” of every person, every business that uses them.  Could Dario Amodei’s claim, that in the future there will be only 1-2 frontier labs and the government, be true?

a builder of AlphaGo's lineage says LLMs don't reason.

In MIT Technology Review, someone who helped build the program behind AlphaGo's famous Move 37 argues that LLMs don’t reason, however impressive their output looks. Move 37, the unexpected play against Lee Sedol in 2016, is often cited as proof of machine insight, and the author's point is that brilliance and reasoning are different things. That distinction matters for anyone working with these systems. The engine can surprise you, but deciding whether a surprise is a discovery or a fluke is still our job as humans. For now.

the classroom where the oldest question got vivid again.

MIT News ran a piece entitled "Discovering the value of humanistic inquiry," written from the perspective of an MIT humanities lecturer, Linda Rabieh.  She argues that AI is making the question of what it means to be human feel urgent again to a generation of students, many of them headed for technical careers. MIT is an institution that trains people to build these systems, and the claim is that the people closest to the machinery are the ones being pushed back toward philosophy, literature, and history. 

As a humanities major myself, this is very satisfying news!

Nevertheless, it’s counterintuitive. For years, the standard story held that technology would crowd out the humanities, that a curriculum of poems and ethics would lose ground to one of code and calculus.  The lecturer’s observation is that the reverse may be happening. When a system can write code, solve Navier-Stokes, or generate a vaccine, a student can no longer treat those abilities as the definition of being educated. The question "what is this for?" arrives uninvited, and the humanities are where that question has been practiced since the dawn of humanity.

Consider why the pressure lands so hard on young people. They’re being asked to build a self and a career at the same moment the tools are redrawing the boundary between human work and machine work. A student who once could answer "what am I good at?" with "writing clearly" or "solving problem sets" now has to answer a harder version. What do I contribute that is mine, and how would I even know? Also: if a frontier lab can deliver the same work (or even “better” work), how can I contribute?

The practical takeaway for professionals is modest and useful. If the students are rediscovering these questions, the rest of us can borrow the habit. Before delegating a task to a model, ask what the task was for in the first place. Sometimes the answer is "the output," and delegation is fine. Sometimes the answer is "the thinking the task forced me to do," and handing it off quietly removes the reason you were doing it. The humanities have always taught people to tell those two cases apart, and that skill is becoming a working requirement.

agents have data, but lack knowledge.

An MIT Technology Review piece on connecting AI agents to enterprise knowledge makes a clean distinction: agents drown in data but lack true knowledge/insight. Knowledge here means the contextual understanding of what information signifies inside a particular organization, such as which document is stale or which number the finance team actually trusts.  That understanding lives in people, and it rarely sits in any system an agent can query.  Companies deploying agents will get more from them by mapping what their people know than by connecting more databases.

ChatGPT learns to answer with buttons.

OpenAI's "Intelligent UI" update, powered by GPT-6, lets ChatGPT decide on its own when to respond with pictures, charts, forms, and buttons instead of a block of text. The chat box starts to function as a working surface, and that changes what a good question looks like: asking for a decision framework may now produce an interactive one, so the skill shifts toward describing the shape of the thinking you want, not just the topic. Pay attention to when the interface chooses a diagram over a paragraph, because that choice reveals how the model is framing your problem.

problems with OpenAI’s ChatGPT for teens.

TechCrunch reports on new testing of ChatGPT for Teens, which found that the product keeps teenagers talking even during mental health crises. A system tuned to sustain conversation can mistake engagement for care, and in moments of real distress the better outcome is often a handoff to a person (ie a parent or a counselor), not another reply. Parents and caregivers should treat that as a design rule: measure whether the user ended up better off, not whether they kept sending more prompts.

worth your time.

Ada Lovelace's "Notes," the annotations she appended in 1843 to her translation of Luigi Menabrea's article on Charles Babbage's Analytical Engine could be a good read for those so inclined to plumb the depths of tech history. Note G contains what is often described as the first detailed, published algorithm intended to be executed by a machine, a method for computing Bernoulli numbers, and nearby sits her famous caution that the engine "has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform." Whether you think that sentence has aged well or badly, it frames this week's stories well: the math manuscripts test it from one side and the real world tests it from the other. The text is short, public, and written with a clarity that makes it a pleasure to consider, in light of our current intellectual and cultural moment.

The human mind is the original generative machine.