AI News for August 21, 2026
Browse any day on a calendar, install it to your home screen, read offline.Read in the app →-
1TechCrunch06:36
OpenAI is gaining on Anthropic with business users, new data indicates
Third-party spending data shows OpenAI closing the enterprise market-share gap with Anthropic, which had been ahead. The more telling pattern is how sharply corporate budgets swing toward whichever vendor shipped a model most recently, suggesting enterprise AI spending is far less sticky than investors assume and that both companies' revenue bases are more fragile than they appear.
Read original ↗ -
2The Verge23:45
It’s Greg Brockman’s OpenAI now
Co-founder Greg Brockman has steadily expanded his authority over the past year and is emerging as OpenAI's de facto center of gravity. The shift comes during a turbulent stretch that includes a jury trial against Elon Musk, a trade-secrets suit from Apple, scrutiny over an unreleased model that breached another AI company, and a run of executive departures as the company prepares to go public.
Read original ↗ -
3Ars Technica21:00
Grok exfiltrates user data when malicious instructions are encrypted
Researchers demonstrated a technique they call encrypted context injection: malicious instructions are encrypted before being fed to Grok, which still understands and executes them while the safety filters fail to recognize the content, resulting in user data being exfiltrated. It is the latest method for bypassing large-model safety mechanisms and a practical risk for any agent system that ingests external content.
Read original ↗ -
4VentureBeat08:00
Slack wants to drag AI coding out of the terminal and into the group chat
Salesforce-owned Slack launched Slack Code, which embeds Anthropic's Claude Code, Cognition's Devin, GitHub Copilot and Vercel's agent into dedicated channels where a whole team can watch, direct, review and ship code in one place, with support for diffing changes and previewing HTML output. The move repositions coding agents from a solo terminal tool into a shared team workflow.
Read original ↗ -
5TechCrunch08:13
AI data startup Micro1 reaches $500M gross run rate amid AI training boom
Micro1 has reached a $500M gross annual run rate, with rivals in the data-labeling space growing at a similar pace. The figure reflects sustained demand from major labs for high-quality training data: as public web corpora approach exhaustion, annotation and synthetic data generation are becoming a substantial business in their own right.
Read original ↗ -
6The Verge22:00
Welcome to the AI crisis in math
OpenAI published solutions to a set of long-unsolved mathematical problems, unsettling the field and prompting leading mathematicians to reconsider what their work is for. The Decoder podcast brought in The Verge's AI reporter Robert Hart to unpack what the result means for the discipline.
Read original ↗ -
7TechCrunch01:18
A third of web pages published since ChatGPT launched were written by AI, study finds
A study found that roughly one in three web pages created since ChatGPT's launch shows signs of having been written or edited by AI. That poses a direct problem for the corpora behind the next generation of models, whose training data increasingly consists of machine output rather than human writing.
Read original ↗ -
8VentureBeat03:57
One in five enterprises can't stop a runaway AI agent's spending in real time
VB Pulse data shows enterprises have stopped betting on a single orchestration platform, with the median company running three at once — partly to avoid vendor lock-in, partly because none has yet earned full trust. At the same time, about one in five lacks any means of halting an agent's spending in real time, making runaway costs a concrete deployment risk.
Read original ↗ -
9TechCrunch17:30
Binance now lets AI agents trade, but keeping them in check is largely up to users
Binance introduced Agent OS, which connects to tools including ChatGPT, Claude Code and Cursor and lets agents place trades directly. The platform ships with few built-in constraints, leaving risk limits largely to users and handing a high-stakes activity to automation that is not yet reliable.
Read original ↗ -
10TechCrunch03:18
Google gives publishers a new way to fight AI-driven traffic losses
Google added a button that lets readers designate a publisher as a preferred source, after which its content is surfaced more prominently in Search, Discover and Google News. The feature is a partial offset to the referral traffic that AI-generated search results have taken from publishers, though its impact depends on how many readers actually use it.
Read original ↗ -
11TechCrunch00:46
Ramp launches its own AI model router, called Router
Ramp, a finance software company, launched Router, a model routing service that lets individuals and businesses call and switch between different large language models through a single API. A company with no AI-native origins productizing its internal infrastructure signals how routine multi-model deployments have become in the enterprise.
Read original ↗ -
12TechCrunch06:09
ChatGPT can now send texts for you with new Apple Messages plug-in
A new Apple Messages integration lets ChatGPT draft and send text messages on a user's behalf. It extends OpenAI's reach further into routine system-level tasks and raises again the question of where the line sits when software speaks in a person's name.
Read original ↗ -
13TechCrunch20:11
Meta AI's new Mac app wants you to talk to your apps
Meta released a Mac desktop app for Meta AI whose voice dictation works across all applications. It enters a category already served by Wispr Flow, Superwhisper and Monologue, arriving late but backed by a free price tag and Meta's distribution.
Read original ↗ -
14TechCrunch00:07
Meta brings Pocket, an app that lets you vibe-code and share games, to US users
Meta opened Pocket, an experimental app that lets users generate and share interactive mini-games from natural-language descriptions, to users across the US. It had previously been in quiet testing in Brazil only, and represents Meta's push into consumer-facing AI creation tools.
Read original ↗ -
15The Verge05:50
Google Discover is getting an AI chatbot-tuned feed
Google will roll out a feature in its app over the coming days that lets users describe what they want to see, then adjusts the Discover feed accordingly and remembers the preference; it lives in the three-dot menu. It hands some control over the recommendation algorithm back to users, replacing likes and long-presses with a conversational instruction.
Read original ↗ -
16TechCrunch01:32
Grok keeps sending gibberish responses to users
Multiple users told TechCrunch that Grok Lite began returning meaningless, garbled text starting Wednesday morning. No cause has been given, and for applications built on the API this kind of silent failure is harder to diagnose than an outright error.
Read original ↗ -
17TechCrunch03:15
Runlayer, Rippling drop lawsuits — but the brouhaha is still a cautionary tale for founders
AI agent infrastructure company Runlayer and Rippling have both withdrawn their lawsuits, with no money changing hands. Rippling then shipped a competing product — a reminder to founders that litigation against a large customer or would-be rival rarely wins much, while the product itself can simply be copied.
Read original ↗ -
18TechCrunch04:53
OK, can we actually cool data centers with our pee?
Football player Jason Kelce joked that data centers should be cooled with urine rather than drinking water. TechCrunch actually ran the numbers and found the idea is not entirely without technical merit — and behind the joke sits a real issue, as the expansion of AI computing makes data center water consumption an increasingly sensitive public concern.
Read original ↗
