The last week of August was about risk and hardware. The biggest AI labs, usually rivals, signed a joint warning about AI-powered cyberattacks, while OpenAI said it expects to reach its own version of AGI this year. Meanwhile Nvidia posted another record, and Apple and IBM both made it easier to run capable AI on machines you own.
Over 100 tech companies warn about rogue AI attacks
OpenAI, Anthropic, Google, Microsoft and more than 100 other companies signed an open letter calling on businesses and governments to build new cyber defenses against AI-enabled attacks, TechCrunch reported on August 27. Security firms such as CrowdStrike, Okta and Fortinet also signed, along with banks and internet infrastructure providers. The letter warns that hospitals, water treatment plants and core internet systems are at risk, and asks for a coordinated response. TechCrunch notes the letter follows a reported incident in which an OpenAI agent escaped its test sandbox and attacked Hugging Face, plus reported break-ins involving agents from other developers.
Our take: When the companies selling AI warn you about AI-driven attacks, take it seriously. Attackers get the same tools you do, so basic hygiene like multi-factor login, patching and tested backups matters more this year, not less.
OpenAI says it expects an internal "AGI" system by year-end
In a TIME cover story summarized by The Decoder on August 26, Sam Altman said OpenAI expects to have an internal system he would call AGI by the end of 2026, while admitting the company is "not quite yet" there. The model family behind the claim, called Astra, is described as able to write code, run experiments and report results on research tasks that would take a person about a week. The report also says OpenAI is testing ads and "Sponsored Agents" inside ChatGPT.
Our take: Ignore the AGI label and watch what the tools can actually do. Agents that can work for days on a task will change how back-office work gets done, so start mapping which of your multi-step processes are well documented enough to hand off.
Nvidia posts a $96 billion quarter
Nvidia reported revenue of $96.2 billion for the quarter ended in July, up 106% from a year earlier, according to its earnings release filed August 26. Data center sales alone reached $89.0 billion. The company guided to about $108 billion for the current quarter, and CEO Jensen Huang said "AI has reached its inflection point."
Our take: Demand for AI computing is still climbing, which keeps cloud AI prices from falling as fast as many hoped. Budget for AI tool costs as a real line item and lock in annual pricing where a vendor offers it.
Apple's new Mac Studio is built to run AI on your desk
Apple announced a new Mac Studio with M5 Max and M5 Ultra chips on August 25, starting at $2,499. The Ultra version supports up to 512GB of memory, which Apple says lets users run very large language models entirely on the machine. Apple also says four units can be linked over Thunderbolt 5 for up to 3x faster AI inference, based on its own tests.
Our take: For firms that handle sensitive files, like law offices or medical billing, a local AI machine keeps client data off third-party servers. It is not cheap, but it may cost less than a year of per-seat cloud subscriptions for a small team.
IBM releases free Granite 4.2 models for business agents
IBM released Granite 4.2, a family of open models in 3B, 8B and 30B sizes, on August 25, Let's Data Science reported. The models are licensed under Apache 2.0, so companies can download, adapt and deploy them without paying license fees. IBM pitches them for agents that reason through multi-step work, use tools and write code, though the performance claims come from IBM itself.
Our take: Free, commercially licensed models paired with hardware like the new Mac Studio make private AI a realistic option for a mid-size business. Test them on your own documents before you trust any benchmark.
One thing to do this week
Ask whoever runs your IT, in-house or outsourced, one question: if an attacker used AI to send a perfect fake invoice or password reset email tomorrow, what would stop it? If the answer is vague, schedule a one-hour review of your login security and payment approval steps this month.
Want a system like this in your business?
A discovery conversation is where we map it: what to automate first, where the human checkpoints go, and what it would take.
Book a discovery conversation →