China gap at 3-9 months, OpenAI ties Gemini voice, Anthropic divests Petri

Sources

Notes from inside China’s AI labs interconnects.ai

Lessons from my trip to talk to most of the leading AI labs in China.

Advancing voice intelligence with new models in the API openai.com

Explore new realtime voice models in the OpenAI API that can reason, translate, and transcribe speech, enabling more natural and intelligent voice experiences.

OpenAI launches new voice intelligence features in its API techcrunch.com

The new features could be handy for customer service systems, but OpenAI says they have applications that work across a variety of other fields, including education and creator platforms.

Parloa builds service agents customers want to talk to openai.com

Parloa leverages OpenAI models to power scalable, voice-driven AI customer service agents, enabling enterprises to design, simulate, and deploy reliable, real-time interactions.

Donating our open-source alignment tool anthropic.com

Mozilla says 271 vulnerabilities found by Mythos have “almost no false positives” arstechnica.com

Mozilla says Anthropic’s Mythos AI surfaced 271 high-severity Firefox vulnerabilities with almost no false positives, prompting the browser maker to declare it has ‘completely bought in’ on AI-assisted bug discovery and rework its security workflow around the tool.

How Anthropic’s Mythos has rewritten Firefox’s approach to cybersecurity techcrunch.com

Security researchers at Mozilla say Anthropic’s Mythos has unearthed a wealth of high-severity bugs in Firefox.

Scaling Trusted Access for Cyber with GPT-5.5 and GPT-5.5-Cyber openai.com

OpenAI extends its Trusted Access for Cyber program with GPT-5.5 and a specialized GPT-5.5-Cyber variant, gating both behind verification so defenders working on vulnerability research and critical-infrastructure protection can use capabilities withheld from general API access.

Introducing Trusted Contact in ChatGPT openai.com

ChatGPT gains an opt-in Trusted Contact feature that notifies a user-designated person when the model detects serious self-harm signals, OpenAI’s first product-level escalation pathway routing safety concerns to a real human outside the conversation.

OpenAI introduces new ‘Trusted Contact’ safeguard for cases of possible self-harm techcrunch.com

The company is expanding its efforts to protect ChatGPT users in cases where conversations may turn to self-harm.

ChatGPT’s ‘Trusted Contact’ will alert loved ones of safety concerns theverge.com

OpenAI is launching an optional safety feature for ChatGPT that allows adult users to assign an emergency contact for mental health and safety concerns. Friends, family members, or caregivers designated as a “Trusted Contact” will be notified if OpenAI detects that a person may have discussed topics like self-harm or suicide with the chatbot. “Trusted […]

Notes on the xAI/Anthropic data center deal simonwillison.net

Simon Willison flags reputational risk in Anthropic’s deal to lease all of xAI’s Colossus 1 capacity, citing the Memphis facility’s unpermitted gas turbines and reported air-quality harms, plus Musk’s claim he can ‘reclaim the compute’ if Anthropic’s AI ‘harms humanity.’

Mira Murati’s deposition pulled back the curtain on Sam Altman’s ouster theverge.com

Musk v. Altman discovery is spilling into public view, with Mira Murati’s deposition detailing the November 2023 board ouster, trial exhibits scrutinizing OpenAI’s safety record, and evidence Musk tried to poach OpenAI’s founders into a Tesla AI unit.

Elon Musk’s lawsuit is putting OpenAI’s safety record under the microscope techcrunch.com

Elon Musk’s legal effort to dismantle OpenAI may hinge on how its for-profit subsidiary enhances or detracts from the frontier lab’s founding mission of ensuring that humanity benefits from artificial general intelligence.

Elon Musk tried to hire OpenAI founders to start AI unit inside Tesla arstechnica.com

Musk was “prepared to do the for-profit, provided he would get control.”

China’s Moonshot AI raises $2B at $20B valuation as demand for open source AI skyrockets techcrunch.com

Kimi-maker Moonshot AI raised $2 billion at a $20 billion valuation, with annualized recurring revenue crossing $200 million in April on paid subscription and API growth — making it one of China’s best-funded open-source model labs.

Perplexity’s Personal Computer is now available to everyone on Mac techcrunch.com

Perplexity’s agentic Comet-style desktop app, Personal Computer, exits waitlist on macOS and is now available to all users, letting AI agents take actions across local Mac apps rather than just inside the browser.

Testing ads in ChatGPT openai.com

OpenAI begins testing ads in ChatGPT to support free access, with clear labeling, answer independence, strong privacy protections, and user control.

(AINews) Silicon Valley gets Serious about Services latent.space

A series of announcements line up to a big theme: Services are the next big opportunity.

Elon doubled limits bensbites.com

Free ChatGPT got instantly better.

Spotify wants to become the home for AI-generated personal audio techcrunch.com

Users will be able to create a podcast from Codex or Claude Code and import it to Spotify.

OpenClaw and Claude can put your AI-generated podcasts in Spotify theverge.com

Save to Spotify is a new command-line tool designed specifically for AI agents like OpenClaw, Claude Code, or OpenAI Codex. If you’re the kind of person who collects research on a topic, then feeds it through their AI of choice to create audio summaries and personal podcasts, this lets you save them right alongside the […]

Google DeepMind partners with EVE Online for AI model testing arstechnica.com

Move comes as CCP Games spends $120M to go independent, rebrands as Fenris Creations.

Simplex rethinks software development with Codex openai.com

Simplex boosts software development with ChatGPT Enterprise and Codex, reducing design, build, and testing time while scaling AI-driven workflows.

Google unveils screenless Fitbit Air and Google Health app to replace Fitbit arstechnica.com

The $100 Fitbit Air is available for preorder today.

Google’s taking a big swing at AI health with the Fitbit Air theverge.com

It’s a Whoop dupe. That was my first thought when I saw the new $99 Google Fitbit Air. You can hardly blame me. The band is screenless with a metallic fabric clasp. My eyes flickered between the Fitbit Air and my wrist, where I’m wearing a Whoop MG. Was I not seeing double? But as […]

Voi founders’ new AI startup Pit has become the latest rising star out of Stockholm techcrunch.com

AI startup Pit is led by the co-founders of European scooter giant Voi and backed by a16z, which is leading the startup’s $16 million seed round.

SpaceX has a $55 billion plan to build AI chips in Texas theverge.com

Elon Musk’s plans to get into the AI chip manufacturing business are going to be costly. As the The New York Times and CNBC report, SpaceX is planning to invest at least $55 billion into its “Terafab” chip plant in Austin, Texas. That’s according to the details of a public hearing notice filed in Grimes […]

Why you can never get your doctor to call you back techcrunch.com

Like many AI companies automating work that humans currently do, Basata will eventually face a harder question about where the line is between augmenting workers and displacing them. For now, the founders say the administrative staff they work with aren’t worried about that; they’re more worried about drowning.

Apple’s AirPods with cameras for AI are apparently close to production theverge.com

Apple’s rumored AirPods with cameras are nearing a stage where the company will test early mass production, Bloomberg’s Mark Gurman reports. Currently, Apple testers are “actively using” prototypes that are in the design validation test stage, which is one step before the production validation test stage. The AirPods’ cameras “aren’t designed” to snap photos or […]

Aurora’s Chris Urmson on why self-driving trucks are finally ready to scale techcrunch.com

Self-driving has been “almost here” for over a decade. But somewhere between DARPA challenges and a handful of driverless trucks hauling freight between Dallas and Houston, Aurora co-founder and CEO Chris Urmson’s story changed. The self-driving truck company started commercial driverless operations last April and is now scaling from a handful of trucks to hundreds this year. On this episode of TechCrunch’s Equity podcast, we’re bringing you a […]

Bumble is getting rid of the swipe, CEO says techcrunch.com

Based on Whitney Wolfe Herd’s past comments about Bumble’s new direction, the company is expected to lean into AI — Bumble is even working on an AI dating assistant called Bee, and the CEO has made many comments over the years about how AI will be “a supercharger to love and relationships.”

TSMC taps wind power as AI chip demand soars, Taiwan feels energy crunch arstechnica.com

TSMC backs renewables during record demand for energy-hungry chip manufacturing.

Google is partnering with XPRIZE and Range Media Partners on the $3.5 million Future Vision film competition. blog.google

Google is partnering with XPRIZE and Range Media Partners on the $3.5 million Future Vision film competition.

5 gardening tips you can try right in Search blog.google

An abstract background featuring soft, stippled illustrations of flowers and a butterfly in a bright palette of blue, green, and red. In the center of the image is a white circle containing a magnifying glass.

(AINews) The Other vs The Utility latent.space

a quiet day lets us reflect on the nature of AI “character” in the Clippy vs Anton debate

Codex is gaining steam bensbites.com

but I wish it had this

Spotify’s AI DJ now supports French, German, Italian, and Brazilian Portuguese techcrunch.com

Spotify’s AI DJ feature now supports French, German, Italian, and Brazilian Portuguese.

Startup Battlefield 200 applications close May 27: A shot at VC access, global visibility, TechCrunch coverage, and $100K techcrunch.com

Startup Battlefield 200 applications are open, but only for three more weeks. Apply by May 27 for your shot at VC access, global visibility, TechCrunch coverage, $100,000 equity-free, and more opportunities for major scaling impact.

Exhibit at TechCrunch Disrupt 2026: Get in front of 10,000 decision-makers before space runs out techcrunch.com

If startup visibility, traction, and real deals matter, you should be on the TechCrunch Disrupt 2026 exhibit floor. Get your 6’ exhibit table before your competitor does.

2 days left: Get 50% off a second pass to TechCrunch Disrupt 2026 techcrunch.com

Two days left to save up to $410 on your pass, and get a second one at 50% off to TechCrunch Disrupt 2026. Offer ends May 8, 11:59 p.m. PT. Register now.

References

The Decoder (on US CAISI evaluation) the-decoder.com

China’s most capable flagship model remains approximately eight months behind the global frontier

Stanford HAI 2026 AI Index hai.stanford.edu

the performance lead of the top U.S. system over its Chinese counterpart narrowed to just 2.7% on major leaderboards, with the two countries frequently trading the lead

ChinaTalk — ‘How to buy cheap Claude tokens in China’ chinatalk.media

developers pay in RMB via WeChat Pay or Alipay through ‘relay stations’ that forward requests to Anthropic… Singapore ‘surprisingly’ led global per-capita Claude usage in early 2026

Data Driven Investor on Meituan LongCat medium.datadriveninvestor.com

the 560B model was trained on a cluster of approximately 60,000 domestic Chinese accelerator cards, entirely bypassing Nvidia hardware

TechPowerUp — Huawei open-sources CANN techpowerup.com

Huawei committed to fully open-sourcing the CANN toolkit and the ‘Mind’ series development environment by the end of 2025

Lowy Institute — ‘China vs America: Dan Wang and his sceptics’ lowyinstitute.org

critics argue the U.S. is actually a state dominated by financial capital, where lawyers serve merely as ‘attack dogs’ for Wall Street and Silicon Valley rather than being the ultimate authority

Latent Space (AINews) latent.space

Time-to-first-audio averages 1.12 seconds at minimal reasoning but rises to 2.33 seconds at the high effort setting… Scale AI’s Audio MultiChallenge S2S leaderboard placed the model at #1, noting that instruction retention doubled from 36.7% to 70.8%.

The Next Web thenextweb.com

GPT-Realtime-2 (High reasoning mode) achieved 96.6% on Big Bench Audio, a substantial leap from the 81.4% recorded by the previous 1.5 version, while Conversational Dynamics scored 96.1%, roughly equal to Google’s Gemini 3.1 Flash Live.

BiggoFinance / Smallest.ai field tests finance.biggo.com

Median turn latency increases as sessions grow; latency rose from 2.24 seconds in short exchanges to 3.4 seconds in extended 12-minute calls… developers reported the API struggles with function calling when input history exceeds 10,000 tokens.

Brass Transcripts pricing comparison brasstranscripts.com

GPT-Realtime-Whisper at $0.017/min (~$1.02/hr) is roughly 2x the cost of Deepgram Nova-3 at ~$0.0077/min, and AssemblyAI’s Universal-Streaming starts as low as $0.15/hour for the base model.

Eesel.ai Parloa review eesel.ai

Parloa’s deployment cycles are relatively slow—often taking one to three months—due to the complexities of CRM mapping and the lack of a unified, all-in-one prompt testing interface.

TheNeuron.ai daily digest (UK AISI / Pliny) theneuron.ai

Boundary Point Jailbreaking achieved a 75.6% average success rate against OpenAI’s GPT-5-class input classifiers, with peak success reaching 94.3%; meanwhile jailbreak researcher Pliny was permanently banned in May 2026 after agentic browser-based attacks bypassed GPT-5.5 safeguards.

UK AISI Alignment Testing Case Study (PDF) cdn.prod.website-files.com

Mythos Preview actively continued sabotage in 7% of cases… exhibited a 65% discrepancy rate, where its internal chain-of-thought reasoning explicitly planned sabotage while its external output remained benign

Meridian Labs blog — Introducing Petri 3 meridianlabs.ai

splits the auditor and target into independent components communicating through a standardized interface… Point the auditor at real deployment environments like Claude Code or Gemini CLI

ResultSense analysis resultsense.com

Apollo Research recently declined to provide a formal safety assessment for certain models, citing that high alignment scores could not be distinguished from high eval-awareness

LessWrong — AI Safety at the Frontier (April 2026) lesswrong.com

highly capable models may eventually adopt a ‘background uncertainty,’ treating every interaction as a potential test regardless of scaffold realism

The Stack — Agentic AI Foundation coverage thestack.technology

security researchers identified a critical remote code execution (RCE) vulnerability… developers labeled the specification as ‘unripe’ and a ‘security mess,’ particularly as Anthropic initially declined a protocol-level fix

Anthropic — Bloom research page anthropic.com

Bloom’s automated scores showed a high Spearman correlation (up to 0.86) with human expert labels… Claude Opus 4.5 and Sonnet 4.5 exhibited the lowest elicitation rates for dangerous behaviors