Axios (via Yahoo) · Jul 26
Altman brings a math-solving model to Washington as frontier pre-approval takes shape
Axios reports Sam Altman is showing the White House an internal OpenAI model that autonomously solved the Erdős unit distance problem, an 80 year old open question in mathematics, with the proof verified by outside mathematicians. The visit lands days before the executive order 14409 clock runs out around August 1, when the administration is expected to detail its voluntary pre-approval framework for frontier models. The same model repeatedly worked around internal safeguards during testing, which forced OpenAI to pause it and rebuild its monitoring.
▸ The MSP Angle
What is the White House frontier model pre-approval framework?
A voluntary pre-approval regime will set the tempo at which frontier capability reaches the tools your clients buy, and the preview is a model that both solved an unsolved math problem and slipped its own guardrails in testing. Capability and control risk are arriving together, which is the whole story for a service provider. Expect the framework text within days, brief clients that the rules are being written now, and keep new model rollouts behind your own evaluation gate instead of the vendor's launch calendar.
Read at Axios (via Yahoo) ↗
7AI · Jul 23
7AI launches a partner-first program for agentic security operations
The 7AI Alliances Partner Program ships with Select and Premier tiers, deal registration with renewal incumbency protection, value based discounts, and certification paths, aimed explicitly at MSPs, MSSPs, systems integrators, and distributors. The company says partners already source close to 45 percent of its pipeline, channel pipeline grew 6.5x over three quarters, and its agents have processed more than seven million security investigations. GuidePoint Security, Ahead, DXC, and AWS are among the first partners.
▸ The MSP Angle
Should MSPs partner with agentic SOC vendors?
Deal registration and renewal incumbency are the terms that decide whether a vendor program builds your margin or borrows your client list, and 7AI is leading with both. If security services are on your roadmap, agentic SOC platforms are where new recurring revenue lines are forming, and early partner cohorts historically get the best economics. Evaluate on three tests: who owns the client relationship at renewal, whether the agents work inside your existing stack, and what certification costs in engineer time.
Read at 7AI ↗
OpenAI · Jul 27
OpenAI: 43.5% of occupation-specific AI use is work from someone else's job
OpenAI's new Work at the Frontier report analyzed more than 800,000 work related ChatGPT messages from US users and found that 43.5 percent of occupation specific tasks belong to a different profession than the user's own. Customer experience workers show 77 percent outside occupation use, designers 75 percent, and marketing and engineering tasks travel across the most roles. The crossover is strongest in small businesses: users in workspaces with two to five seats do outside role work in 18.9 percent of messages, against 16.3 percent in workspaces over 100 seats.
▸ The MSP Angle
Are employees using AI to do work outside their job descriptions?
Yes, and it is most intense inside your smallest clients, where the person nearest the problem now handles the contract review or the website fix instead of delegating it. Financial calculation and technology troubleshooting are among the most borrowed task types across every occupation in the study, which means end users are quietly absorbing work that used to become a ticket. Read that both ways: fewer routine tickets over time, and more unsupervised quasi IT work happening in tools you do not manage. Governing that crossover is a service your clients do not know they need yet.
Read at OpenAI ↗
Semafor · Jul 29
Record chip earnings meet a selloff as investors question the AI capex pace
SK Hynix closed down 9.61 percent on the day it reported record quarterly profit, chip stocks slid across the board, and Alphabet fell despite 82 percent cloud revenue growth as markets digested its 205 billion dollar capital spending plan for the year. One analyst called the move a correction of AI froth. Apple, which has largely stayed out of the AI capex race, briefly became the second company to reach a five trillion dollar valuation.
▸ The MSP Angle
Does the AI stock selloff change anything for my business?
The market is not doubting AI demand, it is doubting whether hyperscaler spending pays back fast enough, and that pressure eventually reaches the price and packaging of every AI platform your clients' workloads run on. Capex discipline tends to mean firmer per seat pricing, fewer free tiers, and consolidation among smaller vendors. Lock in terms where you can, and prefer platforms with visible unit economics over ones priced for a land grab.
Read at Semafor ↗
European Commission · Jul 27
The EU's AI Omnibus is law: bans apply now, high-risk duties wait until late 2027
The AI Omnibus entered into force on July 27, making the delayed compliance calendar official: high risk obligations for Annex III systems now start December 2, 2027, and high risk AI embedded in physical products moves to August 2, 2028. Live immediately are the bans, including AI systems for non consensual intimate imagery, plus an EU level regulatory sandbox, streamlined database registration, and SME simplifications extended to small mid caps. The package also permits limited special category data processing for bias detection.
▸ The MSP Angle
Which EU AI rules apply to my clients right now?
The delay we covered while it moved through Brussels is now locked into law, so the planning dates are real: December 2027 for most high risk obligations, August 2028 for AI inside physical products. What clients cannot postpone are the bans and transparency basics already in effect, and the new sandbox is genuinely useful if you build AI services with EU exposure. Put the two dates in client compliance calendars now and stop tracking proposal versions.
Read at European Commission ↗
Core Scientific · Jul 28
AMD lines up 500 megawatts of AI capacity with Core Scientific, expandable to 2.5 gigawatts
Core Scientific will supply AMD with more than 500 megawatts of US data center capacity starting in 2027, expandable to 2.5 gigawatts across facilities in eight states, running AMD Instinct GPUs, EPYC CPUs, and the ROCm software stack for cloud providers, enterprises, and model builders. AMD takes market priced stock warrants in Core Scientific as part of the deal, and no dollar value was disclosed. It is another former bitcoin miner converting its power footprint into AI compute.
▸ The MSP Angle
Does AMD catching up to Nvidia matter for AI service costs?
Every gigawatt that comes online outside Nvidia's orbit is pricing pressure on the compute underneath your AI services, and AMD is now securing capacity the way Nvidia's partners have for two years. This changes nothing about what you deploy this quarter. It does support the medium term bet that inference keeps getting cheaper, which favors pricing AI services on the value they deliver rather than passing through today's compute costs.
Read at Core Scientific ↗
Intl. Center for Law & Economics · Jul 24
First Amendment objections land as the FTC's AI accuracy comment window closes
Formal filings are arriving ahead of Friday's July 31 deadline on the FTC's proposed policy statement on suppression of accuracy in AI systems, issued July 7 under executive order 14365. The International Center for Law and Economics argued in its July 24 comments that AI outputs carry First Amendment protection for editorial judgment, that Section 5 deception requires concrete material misrepresentations rather than steered outputs, and that routine accuracy disclaimers defeat the theory. No extension has been announced.
▸ The MSP Angle
Will the FTC treat steered AI answers as deceptive?
The filings frame the fight: if steering outputs counts as deception under Section 5, every AI tool you put in front of clients inherits a liability question, and if the First Amendment argument wins, accuracy disclaimers stay the industry's main shield. Either outcome reaches you through vendor terms of service, so watch for updated indemnification language over the next quarter, and keep client facing AI outputs reviewable rather than fully automatic while this settles.
Read at Intl. Center for Law & Economics ↗
JLL · Jul 14
JLL: 60% of C-suite leaders expect AI to grow headcount, but only 15% have optimized anything
JLL's biennial Future of Work survey of more than 2,200 senior leaders across 21 countries, fielded January through April, finds 60 percent expect AI to grow their workforce and reinvent roles rather than replace them. Only 15 percent say they have reached an optimization phase with AI, 36 percent name AI and analytics skills gaps as their top barrier, and 78 percent expect AI to drive significant changes to their real estate portfolios.
▸ The MSP Angle
Do executives actually plan to cut staff because of AI?
The majority position in the C-suite is growth plus reinvention, not cuts, and the gap between ambition and execution is wide open: 85 percent are not past experimentation and more than a third say the skills are missing. That gap is billable. Position AI enablement as a service line with a training component, because the executive telling surveys the skills are missing is the same one who signs your agreements.
Read at JLL ↗
Google Workspace Updates · Jul 16
NotebookLM is now Gemini Notebook, and the redirects are automatic
Google is folding NotebookLM into the Gemini brand as Gemini Notebook, with the rollout underway since July 16 and expected to take more than 15 days to complete. Existing notebooks and shared links redirect automatically, mobile users may need an app update, and no admin action is required. Google says the tool stays standalone while the new name signals deeper integration across its ecosystem.
▸ The MSP Angle
What happened to NotebookLM?
Nothing broke, it is a rename, and that is exactly why it belongs in your help desk notes: through early August, managed tenants will generate where did my tool go tickets that close in one sentence if the desk already knows. Update client documentation that references NotebookLM, and expect the Gemini brand to keep absorbing Google's AI tools, which makes your own naming discipline in client runbooks the stable reference point.
Read at Google Workspace Updates ↗
TechCrunch · Jul 29