AI News

safety fears anthropic openai cheaper ai release a cage of vertical bars with a flat roof

Anthropic, OpenAI Release Cheaper AI Even as Safety Fears Grow

Anthropic released Claude Opus 5.5 and OpenAI released GPT-6 Sol and Luna on 22 September 2026, both cheaper than the models they replace, ten days after the labs’ own leaders called for a slowdown and nine weeks after the Hugging Face breach. This article sets out what Anthropic shipped to answer the safety fears, including rerouting most cyber tasks to Opus 4.8, what OpenAI said instead, why cheaper AI sharpens the risk, and how to read a safety claim attached to a price cut.

Read more
openai app hosting free prototype accelerate plans a rocket body with a cone nose and three fins

OpenAI Introduces Free, Prototype, and Accelerate Plans for AI App Hosting

A new OpenAI Platform onboarding screen spotted on 23 September 2026 offers three plans: Free, Prototype starting at $5 and Accelerate starting at $50. The headline says they are for AI app hosting, but no plan card mentions hosting, and the prices match OpenAI’s existing Tier 1 and Tier 2 thresholds exactly. This article reads the screen line by line, works through what $50 buys in tokens, sets out what OpenAI already hosts for developers, and lists what DevDay on 29 September needs to answer.

Read more
openai creator product patreon execs sam yam a coin tray with four round recesses

OpenAI Nabs Key Patreon Execs Ahead of an Upcoming Announcement

Patreon co-founder and chief technology officer Sam Yam announced on 23 September 2026 that he is joining OpenAI to lead a division called Creator Product, bringing former product head Drew Rowny and former engineering head Shannon Ma with him. The announcement points at OpenAI DevDay on 29 September and at a separate Sam Altman tease, and it lands two months after Patreon cut 93 staff and one month after a 30-feature roadmap. This article separates the confirmed facts from the inference.

Read more
meta smart glasses ai privacy pushback connect a aperture disc with six wedge segments

Meta Leans Into AI and Smart Glasses Amid Privacy Pushback

Mark Zuckerberg opened Meta’s Connect conference in Menlo Park on 23 September 2026 with a keynote built around AI-era products, weeks after the launch of the Muse agentic app and a month after an $18 billion settlement with US state attorneys general over children’s online privacy. Meta and EssilorLuxottica sold more than 7 million smart glasses in 2025 against about 2 million in the two previous years combined, with reported ambitions of 20 million a year. This article separates the product lines, the numbers and the privacy objections that actually hold.

Read more
ai employee ema 77m series b enterprise saas a four stacked discs of decreasing radius

Ema Raises $77M as AI Starts Eating Into Enterprise Software and Services

Ema raised a $77 million Series B on 23 September 2026, led by Bengaluru-based Creaegis with Accel, Section 32 and Prosus increasing their stakes, taking total funding to $140 million on a valuation more than four times the 2024 round. This article separates bookings from revenue, does the arithmetic on the Wipro and Hitachi deployments, explains why outcome-based pricing is what makes the AI employee pitch coherent, and sets out the risks the announcement does not cover.

Read more
muse security flaw meta ai assistant zero day a thick round vault plug with a stepped rim

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

Meta shipped Muse on 8 September 2026 with an undocumented setting, endo_voyager_dictation_endpoint, that any process running as the user could rewrite to capture dictated audio and the token controlling the agent. Patrick Wardle published a working proof of concept on 21 September; Meta hotfixed the Mac app a day later. This article sets out the mechanism stage by stage, separates what was demonstrated from what was described, puts Meta’s account beside the researcher’s, and covers the enterprise visibility problem the patch did not address.

Read more
youtube ai creator tools studio agent a tall stepped tower of three round drums

YouTube Is Building AI Creator Tools That Do Almost Everything for Them

YouTube used Made on YouTube on 23 September 2026 to ship AI creator tools that go well past assistance: a background agent that re-optimises a back catalogue without being asked, generated thumbnails and titles, machine-written sponsor pitches, and three-way video tests that YouTube resolves automatically after seven days if the creator does not. This article separates the eight tasks by who triggers them, does the arithmetic on YouTube’s 40 million A/B experiments, and names the one metric the company declined to publish.

Read more
ai day singapore nvidia southeast asia a stupa dome on a square stepped terrace

At AI Day Singapore, NVIDIA and Partners Showcase AI Advancements Across Southeast Asia

NVIDIA used AI Day Singapore on 22 and 23 September 2026 to publish a regional announcement rather than a product launch: fifteen country-attributed partner organisations across five Southeast Asian countries, one enterprise adopting Vera Rubin, and no new hardware. This article counts the partners by country, supplies the Nemotron 3 parameter sizes the announcement omits, separates research-stage claims from production deployments, adds the Nemotron-SEA-LION v4.8 benchmark gains published four days earlier, and notes that the region’s two largest markets have no named partner at all.

Read more
grok remote control xai desktop app a strongbox with a thick lid and one recessed front panel

xAI Develops Remote Control Feature and Grok Desktop Application

xAI has spent four months building a remote control capability for Grok Build in public without naming it. The evidence is a Computer Hub command hidden from the agent’s own help output, roughly 6.1 megabytes of Rust across eight crates in xAI’s own open-source repository, and a trail of changelog entries running from 3 June to 20 September 2026. A Grok desktop application is referenced three times in those release notes and appears on no download page. This article dates every step from primary sources, measures the shipped code, and sets out what the trust boundary looks like if Grok Build already runs on a machine with credentials on it.

Read more
ai price war anthropic openai cheaper ai models a seesaw plank on a central fulcrum wedge

Anthropic and OpenAI Announce More Powerful (and Cheaper) AI Models

Anthropic released Claude Opus 5.5 and OpenAI added GPT-6 Sol and Luna within hours of each other on 22 September 2026, both cutting prices hard. OpenAI confirmed its new rates are permanent list prices rather than promotions. This article prices the same job on all three models, unpacks why the Anthropic 40 per cent claim is only 20 per cent on the price list, sets out where Opus 5.5 leads on benchmarks and where GPT-6 Astra still beats it, and asks how a price war squares with the industry slowdown both labs chief executives called for two weeks earlier.

Read more
CHAT