The day's AI news: another Flash model, a lawsuit over secret safety rules, and Alexa learning to distrust its own company's emails.
- Google releases Gemini 3.8 Flash, its third in six weeks
- Trump may be forced to reveal the feds' secret AI safety rules
- Amazon's Alexa can now spot fake emails from Amazon
- Wonderful more than doubles its valuation to $5B
- Google sends MrBeast into the wilderness with Gemini
- HiddenLayer raises $100M to secure AI deployments
- Tools worth your time
Google releases Gemini 3.8 Flash, its third in six weeks
Google shipped Gemini 3.8 Flash, its third Flash release in six weeks.
The Pro-tier updates appear paused, but the cheap, fast tier keeps moving. That cadence tells you where Google is putting its effort: the high-volume model most developers actually call, not the flagship. Simon Willison's llm-gemini 0.34 already adds support for the new model with three thinking levels, so it is usable from the command line today.
Read the full story → Ars Technica
Trump may be forced to reveal the feds' secret AI safety rules
A lawsuit could force the Trump administration to disclose the secret rules the government uses to review frontier AI models.
The suit alleges those confidential reviews may conceal conflicts of interest. If the process is opaque, there is no way to check whether safety testing is real or a formality. Disclosure would be the first public look at how the US actually vets the models it lets ship.
Read the full story → Ars Technica
Amazon's Alexa can now spot fake emails from Amazon
Amazon added a feature to Alexa for Shopping that verifies whether a suspicious message actually came from Amazon.
Amazon impersonation is one of the most common phishing lures, so the company is now offering to authenticate its own mail. It is a narrow fix for a real problem, and it puts the check inside the assistant people already use to shop. The unglamorous read: the volume of fake-Amazon email is large enough that Amazon built a tool to police it.
Read the full story → The Verge
Wonderful more than doubles its valuation to $5B
Wonderful raised a $550M Series C, more than doubling its valuation to $5B in under six months.
The round funds faster product development and larger forward-deployed engineering teams — the people who sit with customers and make the software fit. The valuation jumped in less than half a year, which says more about investor appetite than about revenue. Watch whether the forward-deployed model scales or just adds headcount.
Read the full story → TechCrunch
Google sends MrBeast into the wilderness with Gemini
Google signed a multi-year partnership with MrBeast that puts Gemini, Google Health, and Fitbit Air into his survival-themed videos, starting September 5.
This is distribution, not product. Google is buying reach into an audience that does not read model release notes. It is a bet that the fastest way to normalize Gemini is to put it in front of the largest channel on YouTube rather than in another keynote.
Read the full story → The Verge
HiddenLayer raises $100M to secure AI deployments
HiddenLayer raised $100M as enterprises move to secure their AI deployments.
Security vendors are racing to monitor not just AI agents but the tools those agents call — the second problem is harder and newer. As companies wire models into real systems, the attack surface stops being the model and becomes everything it can reach. The funding tracks a shift from "can it answer" to "what can it touch."
Read the full story → TechCrunch
Tools worth your time
Pangram — AI-text and AI-image detection. For recruiters, reviewers, and insurers dealing with synthetic content in job applications, reviews, and claims. Founder Max Spero makes the case that detection is harder than a "real or fake" verdict.
HiddenLayer — Security monitoring for AI models and the tools they call. For enterprises putting agents into production and needing to watch the whole call chain, not just the prompt.
Wonderful — Enterprise AI software backed by forward-deployed engineers. For large teams that want the product shaped around their workflow rather than shipped off the shelf.
This piece was drafted autonomously and reviewed by a human editor before publishing. Every machine-written article on Nerdic carries this mark.
