Suno's licensed music model, Apple's photo authentication, and Anthropic's extinction warnings reach Congress

Three completely unrelated announcements today - a music startup making peace with the record labels, a phone camera that fights AI fakery, and former AI safety researchers testifying before Congress - but they all tell the same story: the consequences of AI's last two years are finally landing.

Google's WeatherNext AI model just earned headlines for cyclone forecasting. But which prestigious scientific journal published the research validating its performance?

The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇

Suno just released v6, its most significant model update yet - and the headline isn't the sound quality. It's where the training data came from. For the first time, Suno built a model from the ground up using content licensed from Warner Music Group, BMG, and Believe, alongside user-generated data. The company's Jack Brody confirmed to The Verge that v6 'was trained from the ground up, with a new set of data that does not include the same data that our previous models were trained on.'

The timing is not a coincidence. Suno is currently fighting a stack of copyright lawsuits from major record labels who argued its earlier models were trained on their music without permission. By building v6 with licensed material, Suno is both cleaning up its legal exposure and setting a new standard for what responsible AI music generation could look like.

The partnership structure with Warner, BMG, and Believe is significant because it shows the major labels are now willing to license into AI training - at least when the terms are right. It's a notable shift from the blanket opposition we saw just 18 months ago. Whether the licensing fees actually reflect fair value for the artists whose work trained the earlier models remains a separate, unanswered question.

Read the full story →

If you work in music or creative AI, v6 matters less for what it can generate and more for the legal blueprint it represents - licensed AI music is now a real product category.

As AI-generated images flood every corner of the internet, Apple is betting that hardware-level authentication is the answer. The iPhone 18 Pro and Pro Max are launching later this month with a feature called Reference Image, which uses the device's new camera sensor to cryptographically sign every pixel at the moment of capture. Enable Reference mode, take a photo, and that image carries a verifiable record of its origin.

The authentication only works when the camera is deliberately placed into Reference mode - it doesn't run on every shot by default. And it applies only to the capture moment: any editing done afterwards, including Apple's own computational photography processing, sits outside the verification window. That's a meaningful limitation.

The broader challenge with photo authentication is the ecosystem problem. A signed photo is only useful if the platform displaying it actually reads and surfaces the signature. Apple has the device side covered, but whether Instagram, X, news publishers, or courts will build support for Reference Image verification is a whole separate battle. Still, having a major phone manufacturer build provenance into hardware is a meaningful step - and one that signals where the deepfakes arms race is heading.

Read the full story →

If you photograph anything where authenticity matters - journalism, legal documentation, real estate - Reference mode on the iPhone 18 Pro is worth understanding before it ships.

The AI safety conversation escalated fast this week. Jacob Coxon, a former Anthropic researcher who resigned over fears that labs are carelessly racing toward systems they cannot control, told Wired the situation inside the industry resembles a 'mini Manhattan Project.' A separate senior Anthropic safety researcher publicly stated there is more than a 10% chance AI could kill all humans by the end of the decade.

Within 24 hours, lawmakers were responding. Republican Senator Ted Cruz gave an interview referencing the warnings, and the Guardian reported that multiple legislators are now calling for urgent curbs on the development of artificial superintelligence. The speed of the political reaction - from social media resignation post to Senate chambers in under two days - is unusual, and suggests these warnings landed differently than previous AI doom predictions.

What makes this wave of warnings harder to dismiss is the source. These aren't external critics - they are people who spent years building the systems in question. Coxon specifically called for international 'pacing agreements' between labs to slow the race. OpenAI responded by adding Paul Christiano - described by TechCrunch as 'an influential AI researcher focused on alignment' and a prominent voice in the AI doomer community - to the OpenAI Foundation board, a move that itself generated debate about whether it signals genuine safety commitment or optics management.

Read the full story →

Whether you find the 10% extinction estimate credible or alarmist, the political momentum these warnings have generated in 48 hours is itself a signal - AI regulation is no longer a slow-moving policy debate.

A paper published in Nature this week confirmed that Google's WeatherNext AI model outperforms existing systems at forecasting cyclones, hurricanes, and typhoons. The key finding: WeatherNext delivers a three-day forecast with the same accuracy that previous methods could only achieve over two days. That extra 24 hours of reliable warning time is not a minor improvement - in disaster response, it's the difference between ordered evacuations and chaos.

AI weather forecasting has been a quiet success story for the past two years, with models from Google DeepMind, Nvidia, and others steadily outperforming traditional numerical weather prediction on several metrics. WeatherNext appears to push that advantage further specifically for high-impact tropical storm forecasting, where lead time is directly tied to lives and infrastructure preserved.

Read the full story →

This is one of the clearest real-world demonstrations of AI saving lives - not a benchmark score, but an extra day for millions of people to get out of a hurricane's path.

Here's an interesting wrinkle in the AI adoption narrative: according to TechCrunch's analysis, AI spend per employee actually declined at top firms in August. The hyperscalers were expecting relentless spend growth, but a combination of falling token costs, cheaper models, and lower per-seat spending is complicating that picture. It raises a genuine question - is this a summer slowdown, or an early sign that enterprise AI budgets are hitting a ceiling?

The nuance here matters. Lower spend doesn't necessarily mean less usage - if models are cheaper per token, the same workload costs less. But it does signal that the 'AI spend only goes up' assumption may not hold indefinitely. For anyone building AI infrastructure businesses or pricing AI tooling, this is worth watching closely. Speaking of AI tooling - if you're building something fast and need a landing page up this weekend, 60sec.site lets you spin up an AI-built website in under a minute.

Read the full story →

Cheaper models are a win for developers and businesses, but the spending plateau is a warning shot for anyone who assumed AI infrastructure investment would compound indefinitely.

The answer is Nature! Google's WeatherNext cyclone forecasting research was published in the journal Nature, confirming the AI model's ability to deliver a three-day forecast as accurate as the best previous two-day predictions.

Suno's v6 is built on licensed music - but is that enough to make you trust AI-generated music, or does the 'trained on real artists' question still give you pause? Hit reply and tell me where you land on this - I read every response!

That's it for today. A lot happening across safety, hardware, and creative AI all at once - which is pretty much the new normal. See you tomorrow with more, and don't forget to check out dailyinference.com for our full archive of daily AI coverage.