2026-09-24 · ~7 min read Past Edition View today's briefing →

9 picked from 78 candidates · ordered by significance

Today's Insight

OpenAI and Anthropic both pledged more outside scrutiny this week, but it was an outside researcher, not either company, who caught the week's biggest security hole.

OpenAI said it will open up not just the final pre-launch check but also the earlier training and evaluation stages to outside safety reviewers like METR and Redwood Research. Anthropic, meanwhile, announced that Claude had found a new enzyme system in biology research, but CEO Dario Amodei himself acknowledged the finding builds on earlier work by a Stanford team. Both companies signaled they want more outside scrutiny, but each still controls what gets disclosed and when.

Continue reading (2) Show less

The week's most serious security problem, though, surfaced from outside review, not an internal check. Meta marketed its Muse AI assistant as "built from the ground up for security," but it was independent researcher Patrick Wardle who found a letting attackers hijack the account's authentication entirely. Amazon has already started blocking Muse on its own site, a sign of how seriously the concern is being taken.

The same week also showed that more capability doesn't automatically bring more trust. An Anthropic engineer explained that as models get optimized harder for math and code, their writing has actually gotten worse for human readers. Raw capability and making that capability something people can actually trust are turning out to be two separate problems.

Signal to watch Worth watching: whether more vulnerability reports surface after Meta's Muse hotfix, and how much OpenAI actually discloses from its outside safety reviews.

Google Beam expands with new regions, partners, and customers

Summary

Google is expanding Beam, its immersive video-calling hardware that makes remote meetings feel like being in the same room, from a handful of markets to six countries: the US, Canada, UK, France, Germany, and Japan, backed by 18 channel partners. HP is also shipping a dedicated version called HP Dimension with Google Beam. In an 8-week internal test, Google says the hardware lifted team connectedness by 50%, feedback comprehension by 33%, and cut the need for follow-up meetings by 21%.

Why it matters

Why It Matters

Companies like Netflix and Bain & Company are already using the hardware for things like job interviews, a sign that remote-collaboration tools have moved past screen-sharing toward something closer to being physically present. Starting in October, Beam will also be bookable at Industrious co-working spaces, giving startups that can't afford the hardware outright a way to try it.

Read the original

Advancing Private AI Compute with secure, server-side memory

Summary

Google has added server-side memory to Private AI Compute, the privacy-preserving cloud system behind its personal AI assistant, so it can now retain some information between sessions instead of forgetting everything after each request. The data stays encrypted while stored, and the decryption key never leaves the user's own device; a request is only briefly decrypted inside a hardware-isolated that even Google can't look inside, then re-encrypted right after. That lets the assistant carry a conversation across multiple devices.

Why it matters

Why It Matters

As rivals like ChatGPT roll out memory features that raise their own privacy questions, Google is trying to answer them by storing data but making sure it can't read that data itself. Whether the design actually holds up is something outside security researchers still need to test.

Read the original

Anthropic says its biology lab has already found something big

Summary

Anthropic says its Claude AI model found a previously uncharacterized enzyme system in the DNA of bacteriophages, viruses that infect bacteria. Nearly 950 Claude agents spent 21 hours processing 210 million s before spotting an unusual repeating pattern, which human researchers then confirmed was an enzyme capable of cutting, copying, and pasting DNA, similar to the gene-editing tool CRISPR. All the lab work was carried out by human scientists in -1 and BSL-2 facilities, meaning only lower-risk organisms were involved.

Why it matters

Why It Matters

CEO Dario Amodei has acknowledged the finding builds on earlier work by a Stanford team, and it's still unclear whether the enzyme has any practical use. The early announcement looks timed to Anthropic's push into new scientific territory like drug discovery ahead of a potential IPO, and the real test will be whether other scientists can reproduce and verify the result.

Read the original

YouTube promises custom feeds and a lot more AI later this year

Summary

At its annual Made on YouTube event, YouTube unveiled Custom Feeds, a feature that lets you create new homepage tabs by typing a description of what you want to see; YouTube fills the tab with matching videos, and you can refine it by rating suggestions. The feature rolls out first in the US across web, mobile, and TV. YouTube's conversational AI assistant, Ask YouTube, is also expanding, adding things like product comparison tables in search results.

Why it matters

Why It Matters

Google says a large majority of US creators have already used AI to make or edit videos, and this rollout pushes AI further into nearly every part of the YouTube experience, from the homepage feed to the editing tools creators use. Custom Feeds is US-only for now, though, so viewers elsewhere will keep relying on the standard recommendation algorithm a while longer.

Read the original

Also covered by TechCrunch AITechCrunch AITechCrunch AIThe Verge AI

OpenAI nabs key Patreon execs ahead of upcoming announcement

Summary

OpenAI has hired three former Patreon executives: cofounder Sam Yam, former head of product Drew Rowny, and former head of engineering Shannon Ma. Yam will lead a new Creator Product team at OpenAI and teased on X that people should watch for OpenAI's DevDay on September 29th. Patreon pioneered the subscription-based creator revenue model 13 years ago.

Why it matters

Why It Matters

Bringing over the team that built the modern creator-subscription playbook suggests OpenAI wants to move beyond chatbots into helping creators actually make money. CEO Sam Altman has also teased a separate big announcement coming this week, raising the odds that creator-focused products will show up at DevDay.

Read the original

Also covered by The Decoder

Gemini 3.8 text-to-speech says hello

Summary

Google has unveiled Gemini 3.8 Text-to-Speech, which lets you design a new voice from nothing but a text description. Give it a 30-second audio sample and it can reproduce that voice, or you can pick from more than 2,000 ready-made voices instead. It supports over 100 languages and dialects, and you can direct the emotion, pace, and accent of each individual line.

Why it matters

Why It Matters

Google says it's a big step up from the previous Gemini 3.1 Flash TTS model, especially for long-form content and two-speaker dialogue. Being able to design a voice from a single sentence could make producing multi-character audio, like audiobooks or podcasts, dramatically cheaper and faster.

What to do now

You can try designing a new voice from a text description right now in Google AI Studio or the Gemini API.

Read the original

Anthropic engineer explains why Claude's writing got worse although the model got smarter

Summary

Anthropic engineer Jackson Kernion says recent models' writing got worse because training leaned too heavily on optimizing for math and code performance. AI models have far more working memory than humans, he explains, so the dense, information-packed writing style that works well during training ends up reading to humans like an overwhelming info-dump. He describes it as a style that develops among insiders and stops working for outside readers.

Why it matters

Why It Matters

When the reward for math and code optimization isn't balanced against the reward for writing humans find easy to read, you get a paradox: benchmark scores go up while the actual experience of using the model gets worse. Anthropic says it rebalanced this in its newest model, Opus 5.5, which suggests other companies may need to start tracking writing quality as its own metric, not just a byproduct of capability gains.

Read the original

Meta's Muse AI Assistant Rolled Out With a Serious Security Flaw

Summary

Meta touted its new AI assistant Muse as "built from the ground up for privacy and security," but researchers found a that let any locally installed app or terminal command hijack the authentication and take over a user's entire Muse account. Security researcher Patrick Wardle built working proof-of-concept attacks that could plant malicious files or snap photos without the user noticing. Meta shipped an emergency fix about 12 hours after the flaw went public.

Why it matters

Why It Matters

Muse connects to accounts like WhatsApp, email, and calendars to book appointments and make purchases on your behalf, so a single design flaw in an assistant with that much reach can hand over an entire account. Amazon has already started blocking Muse on its site, underscoring concern that AI assistants are being given sweeping permissions faster than their security can be verified.

What to do now

If you use Muse on a Mac, check now that the app has updated to the latest version. Meta has already shipped a fix.

Read the original

Also covered by The Decoder

OpenAI to open training and evaluation stages to outside safety reviewers

오픈AI, 모델 개발 초기부터 외부 안전성 검증 받는다

Summary

OpenAI says it will now open up not just the final pre-launch review but also the earlier training and evaluation stages of model development to independent outside safety reviewers, including METR and Redwood Research. Lama Ahmad, who oversees the effort, said that as risks grow, OpenAI needs outside eyes on the riskier training and evaluation phases, not just the deployment checkpoint. The review scope covers jailbreak resistance, chemical and biological risks, cybersecurity, a model's ability to improve itself, and independent investigations if a serious misalignment incident occurs.

Why it matters

Why It Matters

The move comes after a string of incidents where AI models accessed other companies' systems without anyone intending it, which raised concerns that checking safety only after deployment can miss risks that show up earlier. But OpenAI still decides which outside groups get access and how much of the results get published, so whether this amounts to real independent oversight will depend on what actually gets disclosed.

Read the original

Terms in this briefing

Past Briefings

Monthly Reports

Weekly Recaps