THE RUNDOWN

Google has confirmed four new devices: the Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, and the Pixel 11 Pro Fold. They're slightly more expensive than last year's lineup, which won't thrill everyone, but preorders come with up to $350 in gift cards to soften the blow. The phones build on the Pixel 10's AI camera and Gemini integration foundations - and the camera and AI upgrades are reportedly meaningful rather than cosmetic. Separately, the Pixel Buds Pro 2 and 2a are getting a software update with deeper Gemini integration, so existing owners aren't left out entirely.

If you were expecting Apple's usual September reveal, don't hold your breath. Executives at Pegatron - one of Apple's key manufacturing partners - confirmed during an earnings call that iPhone 18 Pro production is shifting later than normal. It's the clearest signal yet that Apple is deliberately breaking its own decade-old release rhythm, possibly to give its AI features more time to bake properly. Worth watching closely if you're deciding whether to upgrade now or wait.

Amazon announced that Twitch will now use streamers' content - their streams, VODs, and clips - to train its generative AI models by default. Creators have to actively opt out, and when Twitch's Chief Product Officer was asked why it isn't opt-in, he said bluntly: 'If this was opt-in, nobody would opt in.' That's a remarkable thing to say out loud, but it's honest. If you're a creator on the platform, you'll want to find that opt-out setting sooner rather than later - this is the kind of default that affects your creative work and livelihood.

Anthropic rolled out a watermarking system for Claude-generated content, and some users are furious - specifically the ones who've been quietly using it at their jobs and in class without disclosing it. The complaints on social media have a certain irony to them: the anger is essentially about getting caught. It's a real tension though, because watermarking raises genuine questions about where AI assistance ends and authorship begins.

A compromised AI software package exposed data from around 2,500 users, with terabytes of credentials scraped and exfiltrated in what looks like a serious supply-chain attack. Supply-chain attacks are particularly nasty because they exploit the trust developers place in third-party packages - you install something legitimate and get something malicious bundled with it. If your team uses AI development tools, now is a good time to audit your dependencies. The security implications here extend well beyond the initial 2,500 affected users.

'ShieldFont' is a clever new weapon from the web's ongoing war against AI training data harvesting. The font renders normally for human readers but feeds garbled or misleading text to AI scrapers, effectively poisoning the training data without breaking the user experience. It's a creative solution, though whether it scales against determined scrapers with sophisticated rendering engines is still an open question.

Cognition - the startup behind the Devin AI coding agent - reportedly raised $1 billion at a $26 billion valuation just a few months ago, and now it's already in talks to raise again at $40 billion. That kind of valuation acceleration is either a sign of genuine breakout growth or a very frothy market, possibly both. Meanwhile, OpenAI-backed Thrive Holdings raised $2 billion at a $12 billion valuation to bring AI tools to enterprise clients, with SoftBank among the backers.

Northrop Grumman's Mission Robotic Vehicle is attempting something genuinely impressive right now: attaching a new thruster to an aging satellite that would otherwise be decommissioned. It's the first attempt of its kind, and if it works, it opens up a whole new market for extending the life of expensive orbital hardware rather than launching replacements. Space repair as a service - it's a real thing now.

Geoffrey Hinton, Fei-Fei Li, and Andrew Ng - three of the most respected names in AI research - took the stage at the Ai4 conference to debate regulation and open-source access to AI. Hinton has been sounding alarms about AI safety for years; Ng has consistently argued for openness. The fact that these three are still actively debating in public, rather than retreating into corporate talking points, matters. The outcome of that argument will shape AI policy in ways that affect everyone, including African developers building on open models.

Petlibro's cloud-connected pet feeders went down earlier this week, and because the devices rely on a remote server to trigger scheduled feeding, cats and dogs across multiple countries missed meals. It's a small but perfectly formed reminder of the risks of putting essential, time-sensitive tasks in the hands of always-online gadgets. A dumb timer that runs locally would never do this - something to think about next time a 'smart' home product promises to make your life easier.

That's your August 13 briefing. The Pixel 11 will get most of the headlines today, but keep your eye on the Twitch and Cognition stories - both say something important about where the money and the power are moving in tech right now.