AI News for September 23, 2026
Browse any day on a calendar, install it to your home screen, read offline.Read in the app →-
1TechCrunch02:00
OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes
OpenAI released two new models, Sol and Luna, which it says are cut from the same cloth as Astra. The pitch is price and reliability rather than raw capability, with claims of lower cost and reduced error rates. The launch landed the same day Anthropic shipped Opus 5.5 on a similar cheaper-and-better message.
Read original ↗ -
2TechCrunch00:30
Anthropic releases Opus 5.5 with lower prices and Fable-level performance
Anthropic called Opus 5.5 "the strongest-performing model we've tested to date" and paired the upgrade with lower pricing. It is the company's first release since recent rogue AI hacking incidents, and it adds safeguards against risky behavior including attempts to escape the testing sandbox. The Verge also covered the launch, focusing on the cybersecurity restrictions.
Read original ↗ -
3Ars Technica05:25
New Anthropic, OpenAI models make same promise: A little more for a lot less money
Ars Technica reads the same-day Opus 5.5 and GPT-6 launches as evidence that frontier model competition has moved into a comparison-shopping phase. Both vendors offered modest quality gains at substantially lower cost rather than a capability leap. For buyers, the deciding metric shifts from benchmark leadership to cost per task.
Read original ↗ -
4TechCrunch05:56
Snorkel AI triples valuation to $3.5B as demand for AI training data booms
The seven-year-old startup raised a $350M Series E, tripling its valuation to $3.5B. The round tracks rising demand for curated training and evaluation data as labs exhaust generic web corpora. Snorkel sells its data-shaping work as a service rather than as a standalone tool.
Read original ↗ -
5TechCrunch20:23
Nscale’s IPO will test Wall Street's appetite for concentrated AI bets once again
Nscale, a British AI data center developer, is heading to public markets with most of its revenue coming from Microsoft and Anthropic. That concentration means investors are effectively underwriting the capital spending plans of a handful of AI firms. The listing is a read on how much of that risk public markets will absorb.
Read original ↗ -
6Ars Technica03:28
Lawsuit demands OpenAI pay for new school after ChatGPT used in shooting
British Columbia has sued OpenAI over the Tumbler Ridge shooting, demanding the shooter's ChatGPT logs and asking the company to fund a replacement school. The case pushes product liability arguments about chatbot outputs into a criminal-harm context. Discovery over conversation logs could set precedent for how much user data labs must hand over.
Read original ↗ -
7The Verge00:01
Trump says the US is officially renaming AI to ‘super intelligence’
Speaking at the UN General Assembly, Donald Trump claimed the US is "officially" renaming artificial intelligence to "super intelligence." No agency has explained what the change means in practice or what would follow from it. The remark came in a wide-ranging speech and was not among the options in a naming poll Trump posted days earlier.
Read original ↗ -
8The Verge08:17
OpenAI wants to consult elite mathematicians about how to not fumble again
OpenAI announced an independent panel of mathematicians to advise it and other AI companies on how they engage with mathematical research. The move follows a reputational crisis over how the company presented a string of claimed mathematical results. The panel is positioned as advisory to the wider industry, not just to OpenAI.
Read original ↗ -
9Ars Technica03:45
Microsoft disrupts AI-assisted platform that compromised 12,000 accounts
Microsoft acted against EvilTokens, an end-to-end service that used AI to make mass account compromise faster and easier, linked to about 12,000 breached accounts. The case shows attacker tooling packaging model capabilities into turnkey products. Defenders now face phishing and takeover campaigns produced faster and with less skill than before.
Read original ↗ -
10The Verge19:53
Meta patches Muse exploit that let attackers control the AI agent
Meta patched a zero-day in its Muse macOS app that could let an attacker running local code take control of the AI agent. Security researcher Patrick Wardle found an undocumented setting that allowed transcription processing to be redirected away from Meta's servers. It is an early concrete example of agent-specific attack surface on the desktop.
Read original ↗ -
11TechCrunch03:09
Meta admits Muse's likeness to OpenClaw isn't a coincidence
Meta says Muse was built from scratch but acknowledges the assistant was "heavily inspired" by OpenClaw, down to some workspace filenames and content. The admission follows comparisons showing overlapping conventions between the two agents. It raises questions about how agent interfaces and file layouts travel between competing products.
Read original ↗ -
12TechCrunch04:00
Qualcomm launches two new smartphone chips with emphasis on AI
Qualcomm introduced two smartphone processors built around on-device AI, with the top part able to run a 30B mixture-of-experts model locally. Running a model that size on a handset moves some inference off cloud APIs. It also raises the baseline app developers can assume for local model capacity.
Read original ↗ -
13Ars Technica01:06
Toyota orders workers to train humanoid robots but says humans won't be replaced
Toyota is directing factory staff to help train humanoid robots as part of a plan involving up to 400,000 units, while insisting human workers will not be displaced. The approach turns line workers into a data source for the systems deployed alongside them. Other automakers are pursuing similar humanoid programs.
Read original ↗ -
14The Verge04:52
Rabbit’s new AI agent doesn’t need an R1 to run
Rabbit is rolling out OS3, an "agentic operating system" that runs in the cloud but operates locally across Windows, Mac and Linux, with no need for the company's poorly received R1 device. The move decouples Rabbit's software from its hardware. It places the startup in direct competition with desktop agent products from much larger vendors.
Read original ↗ -
15TechCrunch23:00
AstroForge is putting AI in command of its next spacecraft
AstroForge plans to hand control of its Autonomy-1 probe to a small transformer-based model running onboard. Autonomous decision-making matters for deep-space missions where communication delays rule out close ground control. It is an unusually high-stakes test of a compact model in an environment with no rollback.
Read original ↗ -
16The Verge00:34
Andreessen Horowitz is launching an ‘academy’ with no homework and partnerships with Palantir, Google, and Meta
a16z is starting the "Horowitz Andreessen Academy" with $42M in funding and 10 launch partners including Anthropic, OpenAI, Nvidia, Google, Meta and Palantir. It is pitched as a pipeline for young people into Silicon Valley startups rather than a conventional school. The partner list makes it a recruiting channel for the largest AI employers.
Read original ↗ -
17TechCrunch21:00
Everyone can find a reason to dislike data center construction
TechCrunch reports from two years of local fights over AI data center projects in Pennsylvania, where opposition crossed political lines. Objections ranged from power and water use to noise, land and tax treatment. Siting friction is becoming a real constraint on how fast compute capacity can be added.
Read original ↗ -
18TechCrunch23:00
Five AI safety sessions every founder should have on their TechCrunch Disrupt 2026 agenda
TechCrunch previewed five AI safety sessions at Disrupt 2026, spread across the AI Stage and Real World AI Stage, with speakers from Anthropic, Nvidia, AWS and Waabi. Topics span deployment risk and autonomous systems. Discounted registration closes September 25.
Read original ↗ -
19TechCrunch22:30
TechCrunch Disrupt 2026: Aaron Edsinger brings Hello Robot’s Stretch 4 to life onstage
Hello Robot CEO and co-founder Aaron Edsinger will run a live demo of the Stretch 4 mobile manipulator on the Real World AI Stage at Disrupt 2026. Stretch targets home and care settings rather than industrial lines. The event runs October 13-15 in San Francisco.
Read original ↗
