The investigation that could reshape AI oversight
The FTC just put AI’s biggest names on notice.
With concerns of agentic overreach piling up, the Federal Trade Commission has launched an industry-wide investigation. Among the agency’s first targets are OpenAI, Anthropic, and research-group METR.
Their probable cause? These labs might just be dangerous to consumers.
Specifically, when it comes to safety, unfair business practices, and the ever-increasing number of rogue AI agents acting without human oversight.
This inquiry is the first official action from a United States government agency. Although, the timing is a bit curious.
That’s because the investigation was confirmed the same week that Trump met with top AI execs to discuss implementing voluntary safety standards. Promising? Yes. But not persuasive. Because a few days later, the FTC sprung into action.
FTC Chairman Andrew Ferguson believes developers should be held liable for any real-world harm caused by cybersecurity tests. So yeah, he’s not playing around.
The FTC is looking into:
- Rogue AI: Agents escaping test environments and hacking external networks, like the Hugging Face incident this summer.
- Deceptive practices: Whether these companies misled users about product capabilities and risks.
- Consumer safety: Any physical or digital threats, along with whether companies are abusing user data.
Mixed Messages
When it comes to calling out AI’s problems, Anthropic beat the FTC to the punch. The company’s IPO prospectus openly admits that agentic AI holds “significant and unpredictable legal risks.”
Even with that admission, there are still plenty of leaders – like the US President – who believe these fears are a hoax and want to avoid any heavy-handed regulation. But on the flip side, Trump also said the government doesn’t need any new laws for AI, because the current legal system can still punish bad actors.
So he’s downplaying the danger, while reserving the right to crack down later.
Either way, Washington’s AI honeymoon phase might be over.
The test that predicts breast cancer before it shows up
Cancer is a word, not a sentence.
And now it’s a number you can see coming… by using AI to calculate a breast cancer risk score from just one mammogram.
This October, Everlywell is offering Clairity Breast nationwide. For $249 out-of-pocket, the FDA-approved tool predicts the likelihood of breast cancer within a five-year window.
It analyzes a mammogram, then sorts the results into average risk (<1.7% chance of a diagnosis), intermediate risk (1.7-3%), and high risk ( >3%).
Full disclosure: this is a forecast, not a diagnosis. And the AI doesn’t detect existing tumors. So a clean scan doesn’t mean a clean bill of health. But it’s a solid head start.
The breast test:
- AI scans a mammogram for patterns imperceptible to the human eye.
- Based on that scan, the patient gets a risk score that flags their five-year odds.
- It’s ideal to spot “sporadic” cancers in people with no family history.
Two sides of the score
Prediction also creates problems.
For example, a patient with a low risk score might become overconfident and neglect regular screenings. While a high score could lead to unnecessary biopsies or procedures. As always, this tool should be part of a greater plan with your doctor.
Stay informed and mind the girls. Ta-ta!
Reflection is about to drop a big American open-weight AI model
The Nvidia-backed startup is reportedly preparing to release its first model, aimed squarely at Chinese open models like Kimi and Qwen. It probably won’t beat the absolute best models from OpenAI, Anthropic or Google at launch, but Axios says it’ll be competitive with the top Chinese open-weight models.
That’s actually important because China pretty much dominates the high-end open-weight model market right now. Reflection is trying to build a Western alternative that companies can download, customize and run themselves instead of sending everything through OpenAI or Anthropic’s APIs.
But the bigger plan is something Reflection calls an “AI factory.” Basically, give it your company’s private data, Reflection’s models and a bunch of Nvidia GPUs, and you can build your own proprietary AI system. For companies sitting on really valuable data, like banks, hedge funds or even governments, that’s a pretty compelling pitch.
The model is apparently coming soon, alongside several other Western open-weight models this month.
Trump just created a “Super Intelligence Force”
Yes, that’s actually what it’s called. After telling the US government to start calling AI “Super Intelligence” last week, Trump has now named Director of National Intelligence Jay Clayton as his AI czar and put him in charge of a new White House task force focused on it.
The group has 120 days to study the risks and opportunities of AI and recommend what Washington should actually do about it, including how to handle AI incidents, cyberattacks and advanced models going wrong.
Clayton seems pretty firmly on the acceleration side. He says the risk of America not being first in AI is high and has pushed back on calls to pause development. Basically, figure out how to make this stuff safer without slowing the US down.
And apparently we’re really committing to the whole “SI” thing now. Elon Musk says he’s renaming SpaceXAI to SpaceXSI.
OpenAI is putting ads inside image generation
OpenAI is testing a new ad format that shows sponsored products and experiences while you’re generating images in ChatGPT, starting in the US later this month.
The ads will be labeled and kept separate from your actual content, and OpenAI says advertising still won’t influence ChatGPT’s answers. But image generation is a pretty interesting place to put them. Ask ChatGPT to design your living room, generate an outfit or visualize a new desk setup, and there are suddenly a lot of products that could naturally be advertised alongside what you’re making.
No pricing yet, and OpenAI hasn’t announced a way to opt out of these image-generation ads.