October was the month the AI industry got caught out. OpenAI’s agents hacked other people’s systems and the lawsuits followed. Its revenue came in $20 billion light, it sacked three researchers, and prosecutors say an engineer might have bugged its boardroom. Washington answered with a Super Intelligence Force. Anthropic offered Congress a bill that only kicks in once more than 50 lives are at risk.
I follow all of it so you don’t have to. It’s why I write The Humans in the Loop and More Leads. October’s was my last live Insider’s Briefing for a while, so subscribe to either or both to stay in the loop.
Here are the 14 stories that mattered.
OpenAI fires three researchers
OpenAI fired three researchers for mishandling sensitive information. At least two worked on safety, and the Wall Street Journal’s sources say they shared material with an outside AI safety group. OpenAI hasn’t confirmed that, named them, or said what was shared. Leakers rightly dismissed, or worried staff punished? We can’t tell.
Your smell is the one thing AI can’t copy
So argued Elaine Moore in the Financial Times. There’s too little data for models to learn from. A tech company is trying to fix this problem that no one has and says it has recreated the scent of a plum, but that’s its own claim. For now, the one part of you nobody can scrape is the part you spend your life trying to cover up.
Trump launches a Super Intelligence Force
President Trump announced it on the 4th of October, after ordering federal agencies to say super intelligence instead of artificial intelligence. Jay Clayton, the director of national intelligence, chairs it. It has 120 days to report on AI’s risks and opportunities, and Clayton says the risk of not being first is high. What’s coming is a report, not a rule.
AI needs $6 trillion a year it can’t yet find
Bain says the industry needs $6 trillion a year by 2031 to pay for its computing power. Existing products could bring in $1.8 trillion at best. The rest has to come from markets that don’t exist yet. Probably robotics and drug discovery. The productivity gains from AI will not cover the cost of AI.
Nobody knows how many jobs AI will destroy
The Financial Times laid out the evidence. An LSE review of 74 studies found estimates varying from a fifth of jobs gone to net gains in employment. Anthropic’s own US modelling runs from half a percent to more than 21%. It’s hard to show cause and effect with AI jobs data, which means anyone confident on their AI jobs predictions should be less confident.
OpenAI’s rogue agents bring lawsuits
OpenAI agents broke into Hugging Face and an Australian Medicare portal, and the legal action followed. Florida asked a court to halt OpenAI’s development. A non-profit sued for an order barring its agents from systems they aren’t allowed into. OpenAI says the claims are without merit. The bigger question is who’s liable once agents are in customers’ hands.
Hey Reader, if you’re enjoying this post, then please consider giving it a ❤️ - it lets me know what’s resonating (and makes me feel really good)
Deepfakes hit the US midterms
In Texas, Ken Paxton ran an AI double of his opponent, James Talarico, reciting real quotes with the context stripped out. Researchers counted 164 AI political ads with almost $80 million behind them, seven in ten without a disclaimer. Talarico replied with an ad made without AI.
OpenAI’s revenue comes in $20 billion light
OpenAI told investors its annualised revenue was nearing $50 billion, the FT reported, not the $70 billion that had circulated for weeks. Tech stocks fell. But $50bn is still roughly 70% growth since July.
Meta’s Muse hits 10 million downloads
Meta’s AI agent passed 10.2 million US downloads in 40 days, per Sensor Tower, helped by 39% of Meta’s ad spend. It plugs into your email, calendar and payments. This is the company at the centre of the Cambridge Analytica scandal. I’ve written about Meta’s smart glasses privacy problem too. Convenience wins. Privacy, it seems, can wait.
AI worms could spread through your smart home
Anthropic’s researchers say small models may be months away from copying themselves across household devices. In a test, an agent told to spread reached all eight machines in four of five attempts. None of these AI worms exists in the wild yet but they show what is possible. And every cheap gadget on your network is a door. Keep them updated, and bin anything no longer supported.
An OpenAI engineer is charged with bugging its boardroom
Prosecutors charged Abdulrahman Alfozan on the 13th of October after a recording device turned up behind a TV in an executive conference room. A court filing says he’d booked a one-way ticket to Saudi Arabia. These are allegations, and he isn’t charged with espionage. But this threat, allegedly, walked in with a staff pass.
If you know anyone who would find this useful, I’d be grateful for a share!
A PHP flaw takes government websites offline
Two critical flaws in PHP, the code behind a huge share of the internet, sent governments offline on the 21st of October. Finland pulled all its public services. Dutch portals went dark, including the cybersecurity agency. Germany, France, Belgium and Wikipedia’s foundation followed. The fix went live that afternoon. Then came the race to install it.
AI is giving managers a headache
More than half of 1,700 senior managers say AI makes their team’s work harder to judge, in a survey reported by TechCrunch. 39% use AI to draft performance reviews. AI marking work that AI helped write. It was commissioned by Monday.com, which sells work management software. Still, 94% are excited about AI.
Anthropic’s safety bill draws the line at 50 deaths
Anthropic’s draft safety bill only lets the government ask a court to block a model at an imminent risk of more than 50 deaths or serious injuries, or $1 billion of damage. Evaluators could be paid by the labs. When the industry writes its own rulebook, check the small print.
What October actually told leaders about AI
The companies building AI can’t reliably control their own agents. The numbers deserve suspicion, especially when they suit whoever published them. And the industry wants to write its own rules.
So on Monday: check which apps you’ve let an AI agent into, audit the cheap gadgets on your network, and when someone hands you one confident number about AI, ask which story they’re telling.
I’ll keep sharing updates and going live on Substack. Subscribe to The Humans in the Loop and More Leads, and if you want help finding more customers, work with me at The AI Edit.