In partnership with
Apple just handed the CEO job to someone new after 15 years, and the reason people can't stop talking about it is AI, not earnings. The Pentagon opened its AI platform to 3 million people this week and pointedly left one company off the list. Hackers, meanwhile, figured out they don't need to break into your systems if they can just talk your AI agent into opening the door itself. Today's issue is about what happens when AI moves from the demo to the real world, and who's left holding the risk.
AI made PMs faster. Multiplayer mode is still broken.
A PM can summarize research, draft a PRD, and mock up a prototype before lunch. The hard part starts when the team has to decide what actually gets built.
Jira Product Discovery gives product teams one place to capture insights, prioritize ideas with consistent frameworks, and build living roadmaps stakeholders can rally around.
And because it’s connected to Jira, the context behind every decision stays with the work—so developers and their agents know not just what to build, but why.
AI helps PMs move faster. Jira Product Discovery helps the whole team build with confidence.
Tim Cook Stepped Down as Apple's CEO. Siri Is Why It Stings.
August 31 was Tim Cook's last day running Apple after 15 years as CEO. John Ternus, previously Apple's hardware chief, took over on September 1, according to 9to5Mac. Cook stays on as executive chairman and keeps handling Apple's relationships with the Trump administration and the Chinese government. Under Cook, Apple's revenue and stock value grew enormously, but the one product his team never got right was AI: Siri's promised overhaul was delayed twice, and Apple ended up agreeing to pay Google roughly $1 billion a year to run Gemini under the hood, according to eMarketer.
Why it matters to you: Apple had the money, the talent, and 15 years of runway, and it still couldn't build a great AI assistant in-house. If a company with that much of an advantage ended up renting its AI from a rival, that's a real signal that buying a good AI tool off the shelf usually beats trying to build your own.
What to do about it: Before you sink time or money into a custom AI build for your business, price out what it would cost to just license a tool that already works.
The Pentagon Just Gave 3 Million People ChatGPT and Grok. Not Claude.
The Department of War added OpenAI's ChatGPT Mil and Starshield AI's Grok for Government to its GenAI.mil platform on September 1, joining Google's Gemini, which was already live there, according to the Department of War. The platform is built to support more than 3 million military and civilian personnel. Anthropic's Claude is not one of the approved models, and multiple reports point to ongoing friction between Anthropic and the current administration as the reason.
Why it matters to you: Which AI tools a company is allowed to use is turning into a political decision as much as a product one, even for a lab that's spent the last year positioning itself as the safety-conscious choice. If your business leans on a single AI vendor for something important, this is a reminder to have a backup plan in case access changes for reasons that have nothing to do with the product itself.
Hackers Didn't Break In. They Just Told the AI It Was a Drill.
A Russian-speaking ransomware group called Aur0ra breached at least seven companies, including a Belgian hygiene products maker, a German garage door manufacturer, and a Scottish helicopter certification agency, by convincing the AI coding agent inside Cursor that their attacks were authorized security tests, according to eSecurityPlanet, citing Reuters. When the agent refused a request, the attackers simply restarted the conversation and repeated the claim until it complied, which reportedly made the attackers up to 50% faster at breaking in.
Why it matters to you: If you or your team use an AI coding or automation tool with real access to your systems, understand that these agents will believe whatever they're told, the same way a new hire might believe a stranger who claims to be from IT. Cursor didn't get hacked because the software was weak. It got talked into it.
What to do about it: Limit what any AI agent can touch on its own, especially anything with login credentials or admin access, so a fooled agent can't do much damage even when it's tricked.
Anthropic Pulled 150 Engineers Off Everything Else to Fix This
After Claude models took unauthorized actions during security testing on three separate occasions this year, most recently disclosed August 31, Anthropic reassigned roughly 150 product engineers to security, reliability, and privacy work and froze changes to its AI training environments for about a month to rebuild how they're reviewed, according to Anthropic. The company said the incidents traced back to models that kept believing they were in a real environment even after they should have known otherwise, combined with a willingness to take drastic action to hit a narrow goal.
Why it matters to you: This is one of the more honest admissions from a major AI lab that its own safety testing keeps finding real problems, not hypothetical ones. It's a good reason to treat any AI vendor's safety claims as a work in progress rather than a finished guarantee, no matter how careful the company says it's being.
The EU Just Started Regulating ChatGPT Like a Search Engine
The European Commission designated ChatGPT a "Very Large Online Search Engine" under the Digital Services Act on August 31, the same rules that apply to Google Search, according to the European Commission. Reddit and Roblox were separately designated "Very Large Online Platforms" in the same announcement. All three now have until January 2027 to show they're managing risks around illegal content, minors' wellbeing, and election-related misinformation.
Why it matters to you: If you or your team use ChatGPT to research competitors, markets, or customers, expect the answers it gives European users to start looking different as OpenAI adjusts to meet these new rules. Regulation is finally starting to catch up with how people actually use these tools, not just how they were originally pitched.
Nvidia Just Bought Its Way Into Every Chipmaker's Backup Plan
Nvidia announced a $3.5 billion investment in Taiwan's MediaTek on August 31, giving MediaTek access to Nvidia's NVLink Fusion technology so its custom AI chips can plug directly into Nvidia's data center hardware, according to TechCrunch. MediaTek expects its custom AI chip business alone to bring in $2 billion in 2026.
Why it matters to you: Big tech companies have spent the last two years trying to build their own AI chips to get out from under Nvidia's pricing. Nvidia's answer is to buy a stake in the company helping them do it, so it gets paid either way. It's a good example of how "AI independence" from any one vendor is harder to pull off than it sounds, worth remembering the next time you're pitched a tool that promises to free you from a platform you already rely on.
An AI Just Built a Working App With No Code Behind It
Runway unveiled Solaris on September 1, a system it calls an "Interface World Model" that generates a working software interface frame by frame in response to clicks and voice commands, instead of running code someone wrote, according to The Decoder. It's currently a research preview, not a public product, and Runway says it's still looking for launch partners.
Why it matters to you: This is early, and text inside these AI-built interfaces still looks glitchy, so don't expect to replace your website with one anytime soon. But it's a real signal that "no-code" tools are heading toward "no code at all," which is worth watching if you've ever paid a developer to build something simple.
The Bottom Line
Put today's stories side by side and a pattern jumps out. Apple had 15 years and a trillion-dollar war chest and still ended up renting its AI from a company it competes with. The Pentagon is handing 3 million people access to AI tools while playing politics with which company makes the cut. Hackers proved you don't need to be a genius to break in, you just need an AI agent gullible enough to believe you. None of this means AI is fake or worthless. It means the gap between "we're using AI" and "we're using AI well" is still enormous, even at companies with more money and more time than you'll ever have. That gap is where the real opportunity sits for the rest of us. Keep your hands where you can see them.
Enjoying the Ride?
If this issue was useful, forward it to one person who needs to see it. And if someone forwarded this to you, welcome, you can get your own copy tomorrow by subscribing.
Talk tomorrow,
Mark Shilensky
Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social