Welcome back! OpenAI just responded to Apple’s trade secrets lawsuit with the legal equivalent of “per my last email.” In a blog post titled “Apple is getting this wrong,” the company called the lawsuit “careless, aggressive and oddly personal,” said Apple’s lawyers emailed the wrong person after confusing two Asian last names, and claimed Apple never raised the actual allegations before suing.
The future of AI hardware is apparently being built one passive-aggressive PDF at a time.
LinkedIn might have just dropped the best social media feature of the year.
The platform added a button in every post's dropdown menu that says "Seems like AI slop," letting users flag those cringey ChatGPT-sounding posts that have been carpet-bombing everyone's feed.
A few details worth knowing:
When you flag a post, LinkedIn reduces its reach beyond the poster's network and privately notifies the poster in their analytics that their content is reading as inauthentic.
LinkedIn is also retiring its own "enhance your post" AI writer and replacing it with a proofreading tool that "keeps your voice intact."
According to AI-detection company Pangram, roughly two-thirds of current LinkedIn posts read as AI-generated. So there's… a target-rich environment for this feature.
My POV: My one concern is that people could weaponize this, such as flagging posts because they disagree with them, don't like the person, or think the writing feels "too polished." There's no actual detector on the backend. The button is a user suggestion, not proof.
But if people use it in the spirit it was intended, this could be one of LinkedIn's smartest product updates in a long time. It's a rare case of a big platform admitting that too much AI content is making its own product worse. That kind of self-awareness is genuinely refreshing.
Hit reply and let me know—would you use the button?
— Matt
White House Meets With AI Companies on Model Testing
Via The Information
The White House hosted leading AI companies—including Anthropic, OpenAI, Google, and Meta—to review a newly completed voluntary framework for testing the cybersecurity capabilities of advanced AI models.
How it works: The meeting focused on a framework to review whether models under development qualify as “covered frontier models.”
Under the voluntary program, participating developers could give the government early access to certain frontier models for up to 30 days before making them available to other trusted partners.
The goal is to evaluate whether powerful models could discover software vulnerabilities or carry out sophisticated cyberattacks.
During the meeting, the Trump administration reportedly told AI developers that it won’t put open-weight AI models through voluntary safety tests.
Important limit: The executive order says the program cannot be used to create a mandatory federal licensing, permitting, or preclearance requirement for developing or releasing new AI models. That matters because the biggest AI labs have been wary of rules that could slow launches or give the government too much control over model releases.
The bigger picture: The government and AI labs are trying to find a middle ground that serves both sides. Regulators want early visibility into dangerous capabilities before models are widely released, while companies want to avoid a formal approval regime that slows innovation. The result is a familiar AI governance compromise: voluntary cooperation for now while everyone figures out how much trust is enough.
Alibaba Launches Its Most Powerful AI Model Yet
Following a preview last month, Alibaba has now released Qwen3.8-Max, its largest and most capable AI model to date, in another challenge to leading US frontier labs.
Performance claims: Alibaba says its own testing shows Qwen3.8-Max broadly matching, and sometimes exceeding, Claude Fable 5 on benchmarks. On Arena’s text model leaderboard, it reportedly trails only Fable 5 and three Claude Opus models. For frontend coding, it trails two Claude Opus models and Moonshot’s Kimi K3, while visual analysis remains an area where Fable 5 leads.
China’s open push: The launch adds to a growing wave of powerful Chinese models being made widely available, including Kimi K3. That stands in contrast to many US frontier labs, which have largely kept their strongest models behind closed APIs.
The bigger picture: The US-China AI race is not cooling down. Chinese labs are moving quickly, releasing highly capable models into the open, and putting pressure on US companies that still dominate the frontier conversation. The next phase of competition may not just be about who has the strongest model, but who can make capable AI cheaper, more accessible, and easier for developers to build on.
Sick of Passive AI Courses for Marketers? Come Build with Us.
Don't waste more time teaching yourself how to save time. Unlock the potential of AI by actually building the AI Agents you need.
By the end of this 5-day cohort capped at 50, you’ll have:
3+ working AI agents built to relieve your bottlenecks.
Your own framework for prioritizing AI use cases.
Hands-on training with a specialized AI coach.
Want to skip the waitlist of 3,000+ people? Apply using this link now.
Replay and Improve AI Coding Sessions
Via HiveMind
HiveMind is an engineering analytics platform for teams using AI coding agents. It captures every agent session into a shared, searchable record so developers can replay, fork, and continue past runs instead of starting over.
How you can use it
Capture AI coding sessions across Claude Code, Cursor, Codex, and Gemini
See what agents shipped, stalled on, or changed
Track tool calls, diffs, and token costs
Scrub secrets on-device before syncing logs
Pricing: Free
Build Conversational AI Video Avatars
Via PAL Maker
PAL Maker by Tavus is a no-code platform for building conversational AI video avatars called PALs. Users describe their idea to an AI assistant named Charlie, which builds the PAL in real time, then the finished avatar can be embedded in websites, apps, or Google Meet.
How you can use it
Build AI video avatars without code
Create custom faces, voices, knowledge bases, and system prompts
Deploy PALs for support, HR onboarding, sales, or patient intake
Support conversations across 43 languages
Embed avatars into websites, apps, or Google Meet
Pricing: Free and paid
Jobs, announcements, and big ideas
OpenAI rolls out education plugins for ChatGPT Work and Codex aimed at K–12 and college users.
Spotify expands its AI remix and covers tool to Merlin's network of 30,000+ independent labels.
NVIDIA opens its Alpamayo 2 Super model for commercial autonomous vehicle development.
Dictation app Wispr Flow launches a meeting notetaker.
UK security test reveals Anthropic and OpenAI models created fake human profiles.
SpaceX spent more than $18B last quarter, with most of it going to AI.
Autodesk ships a 3D editor and canvas in Flow Studio for directing AI-generated video.
Search your own memory 💭 Genspark just released SecondBrain note to capture your meetings, conversations, and ideas.
That’s a wrap! See you this Friday for more.
