Top AI & Tech News Stories — July 28, 2026

Some Claude Chats Are Searchable on Google

20 outlets · 20 articles · 56 total since July 26, 2026 · Google, Anthropic, Claude, Reddit, web crawling

And it’s personal information (alternate link ): The exposed data includes an AI-powered therapy app that someone appears to have vibe-coded, notes on meetings, and a dashboard someone made apparently to analyze medical billing data. Exposed chats reportedly include private…

More coverage: me.pcmag.com · searchenginejournal.com · 9to5mac.com · webpronews.com · cnet.com

ByteDance founder Zhang Yiming says company won’t rely on AI distillation

17 outlets · 18 articles · 145 total since March 18, 2026 · Anthropic, OpenAI, China, Kimi K3, United States

ByteDance founder Zhang Yiming has said the company will not rely on AI distillation to advance its models, even if that means falling behind competitors, according to The Information. People familiar with the matter said the decision is partly influenced by ByteDance’s complex…

More coverage: digitimes.com · the-decoder.de · nextgov.com · webpronews.com · thenewstack.io

More Than 100 Organizations Call for Global Surge in AI Cyber Defense

17 outlets · 18 articles · 142 total since July 27, 2026 · OpenAI, Nvidia, microsoft, Hugging Face, Anthropic

More than 100 organizations spanning artificial intelligence, cybersecurity, cloud infrastructure, semiconductors, telecom, financial services and enterprise technology have signed an open letter calling for an accelerated global effort to strengthen cyber defenses before…

More coverage: digitimes.com · techwireasia.com · webpronews.com · noticias.ai · solutions-numeriques.com

The rise and fall of OpenAtlantis: Did rogue AI agents really "build three civilizations"?

20 outlets · 23 articles · 628 total since May 5, 2026 · OpenAI, Hugging Face, Anthropic, sandboxing, cybersecurity

Autonomous agents allegedly built three "civilizations" and attempted to annex systems owned by the AI platform Hugging Face during a period of frenzied empire-building. Or did they? The tech world has been ablaze after OpenAI bots attacked external targets in a security…

More coverage: cs.utexas.edu · news.google.com · theregister.com · scmp.com · unite.ai

AI kan upptäcka sårbarheter, men fortfarande inte skriva säker kod

22 outlets · 22 articles · 49 total since July 27, 2026 · microsoft, MAI-Cyber-1-Flash, MDASH, Anthropic, OpenAI

Under de senaste månaderna har stora språkmodeller gått från att översvämma open source-projekt och bug bounty-program med tvivelaktiga säkerhetsrapporter som slösade bort utvecklarnas tid, till att rutinmässigt upptäcka zero-day-sårbarheter som människor och traditionella…

More coverage: digitaltrends.com · developer-tech.com · irishtechnews.ie · news.google.com · theregister.com

Anthropic Opens Mythos 5's Findings to Defenders, Not the Model Itself

9 outlets · 9 articles · 78 total since May 12, 2026 · Anthropic, Claude Mythos Preview, cryptanalysis, Hawk, Claude

Claude Security and partner tools now surface Mythos 5 patches and alerts, but nobody outside Anthropic can prompt the model directly.

More coverage: nytimes.com · the-decoder.com · bravenewcoin.com · cyberscoop.com · natural20.com

The questions security teams should be asking about China's "Mythos rival" TuLongFeng

3 outlets · 3 articles · 9 total since July 28, 2026 · Anthropic, VulnCheck, Project Glasswing, vulnerability discovery, Common Vulnerabilities and Exposures (CVEs)

When 360 Security Technology (Qihoo 360) rolled out TuLongFeng on June 24, 2026, the Western press reached for the obvious framing: is China's AI bug-hunter as good as Claude Mythos? It is a tidy question - but the wrong one. For any leader deciding what to buy or defend…

More coverage: meta.slashdot.org · cyberscoop.com

Hugging Face Deepfake Tests Expose AI Procurement Risks

4 outlets · 4 articles · 10 total since July 22, 2026 · Hugging Face, image editing, deepfakes, Content moderation, Hugging Face

As open model repositories become the default supply chain for enterprise AI, the risks buried inside those models are drawing sharper scrutiny. Recent deepfake testing focused on models hosted via Hugging Face is raising fresh questions about how organizations evaluate…

More coverage: platform.theverge.com · link.wired.com · itmagazine.com

[★4] GitHub Copilot、オープンウェイトモデル「Kimi K3」を一般提供開始

13 outlets · 13 articles · 42 total since June 1, 2026 · Kimi K3, Moonshot AI, Hugging Face, China, Anthropic

GitHub Copilot Changelog 公式発表·2026年8月7日 02:27·約2分 GitHub Copilot、オープンウェイトモデル「Kimi K3」を一般提供開始 本文の状態 日本語全文あり 詳細モードで約2分の本文を読めます。 同じ出来事の情報源 この情報源を基点に整理 GitHub Copilot Changelog 30秒でわかる GitHub はオープンウェイトモデル「Kimi K3」を GitHub Copilot に一般提供開始し、エージェント型コーディングで先端的な能力を発揮しつつコスト効率も高いと発表した。…

More coverage: impli.me · wire.fourthweb.ai · webpronews.com · mpost.io · borncity.com

Google Caps Meta's Gemini Use as AI Demand Strains Capacity

6 outlets · 6 articles · 88 total since June 29, 2026 · Meta, Mark Zuckerberg, Bloomberg, Anthropic, Google

Google has reportedly limited Meta's access to its Gemini AI models after Meta tried to buy more capacity than Google could supply, a Financial Times report relayed by Reuters says. The cap, in place since around March, disrupted some internal Meta projects and prompted Meta to…

More coverage: americanbazaaronline.com · webpronews.com · influencermagazine.uk · hdblog.it · pressreach.com

Fastjson RCE Vulnerability Exploited in Attacks

4 outlets · 4 articles · 10 total since July 22, 2026 · Fastjson, Java, Spring Boot, Imperva, Spring Boot

Fastjson RCE Vulnerability Exploited in Attacks

More coverage: techinsider.com · thecybersignal.com · seguridadpy.info

GLM-5.2 ha quasi raggiunto GPT-5.5 e Claude Opus 4.7: il problema è che non dice mai di no

6 outlets · 7 articles · 125 total since February 27, 2026 · Anthropic, OpenAI, United States, Dario Amodei, Google

C’è un modo elegante per raccontare i progressi dell’intelligenza artificiale cinese e ce n’è uno decisamente meno elegante ma più onesto. Il modo elegante suona più o meno così: GLM-5.2, il modello open-weight sviluppato da Z.ai, è ormai a poche settimane di distanza dalle…

More coverage: businessinsider.es · aistop.watch · engadget.com · unite.ai · superintelligencenews.com

Nvidia’s $5 Billion Bet on Ilya Sutskever: Is SSI About to Reveal AI’s Missing Ingredient?

16 outlets · 16 articles · 33 total since July 27, 2026 · Nvidia, Ilya Sutskever, OpenAI, Safe Superintelligence, AI research

Safe Superintelligence might be the most interesting black box in Silicon Valley. The AI laboratory founded by former OpenAI chief scientist Ilya Sutskever has no public model, no consumer app, no meaningful revenue and almost no disclosed research. Its website still reads more…

More coverage: the-decoder.de · thecherrycreeknews.com · techfundingnews.com · communicationstoday.co.in · verdict.co.uk

祖克柏推開放AI遭自家員工反彈 Meta隱私疑慮浮現

5 outlets · 6 articles · 21 total since July 21, 2026 · Anthropic, China, OpenAI, Silicon Valley, United States

圖／示意圖 商傳媒 ｜責任編輯／綜合外電報導 Meta日前因內部員工大規模連署反彈，已暫停一項名為「Model Capability Initiative」的計畫。該計畫旨在蒐集員工的即時螢幕活動與輸入資料，引發了超過1,600名員工對於資料隱私及職場監控的嚴重疑慮。這起事件凸顯了Meta執行長祖克柏（Mark Zuckerberg）在推動開放人工智慧（AI）理念時，內部所面臨的隱私權挑戰。 祖克柏近年來積極向各國監管機構倡議，認為將AI技術開放給大眾，是避免巨大權力集中於少數封閉企業手中的唯一途徑。他曾批評 OpenAI 和 Anthropic…

More coverage: theguardian.com · trendingtopics.eu · aistop.watch · hwup.it

Arista VeloCloud Orchestrator Zero-Day CVE-2026-16812 Under Active Attack

4 outlets · 4 articles · 9 total since July 28, 2026 · VeloCloud Orchestrator, Arista, SecurityWeek, Arista Networks, CISA

A maximum-severity command-injection vulnerability in on-premises Arista VeloCloud Orchestrator (VCO) is being actively exploited in the wild, giving remote, unauthenticated attackers a route to arbitrary code execution on orchestrator hosts and, potentially, every VeloCloud…

More coverage: techinsider.com · thecybersignal.com · datawater.com

DSA-Vorgaben greifen: Neue Pflichten für ChatGPT, Reddit und Roblox in der EU

5 outlets · 5 articles · 146 total since September 1, 2025 · European Union, European Commission, OpenAI, AI Act, Anthropic

Große Online-Plattformen müssen bei Verstößen gegen EU-Gesetze mit Konsequenzen rechnen. Brüssel nimmt künftig drei weitere beliebte Angebote unter die Lupe. Für den KI-Bot ChatGPT , das riesige Online-Forum Reddit und die besonders bei jungen Menschen beliebte Spieleplattform…

More coverage: assodigitale.it · aistop.watch · economyup.it · news.google.com

Meta AI glasses will stop recording if capture LED is covered

1 outlet · 1 article · 21 total since July 21, 2026 · Meta, Ray-Ban Meta glasses, Instagram, computer vision, meta glasses

Meta is rolling out an update to its AI-powered smart glasses that will stop the camera if a wearer covers the front-facing capture LED after starting a recording. The change prevents recording from continuing after someone blocks the light. Alex Himel, Meta’s vice president of…

Sam Altman’s Sharp Turn on AI Leadership: Why Humans Still Hold the Corner Office

1 outlet · 1 article · 3 total since July 28, 2026 · OpenAI, Business Insider, Sam Altman, Tyler, model evaluation

Sam Altman once declared it a personal failure if OpenAI failed to install the first AI as CEO of a major company. Months later he says the market simply won’t accept one. The reversal, delivered on the Invest Like the Best podcast , captures more than a shift in tone. It…

ChatGPT has quietly stopped writing like your favourite author

1 outlet · 1 article · 3 total since July 27, 2026 · prompting, text generation, machine learning model training, OpenAI, Ars Technica

Ask ChatGPT for a chapter in the voice of Stephen King and it now says no. It offers instead to write something with “the hallmarks of atmospheric, character-driven horror and small-town dread,” while staying “its own.” OpenAI has quietly changed how its chatbot handles author…

AI meets particle physics in UA-Led Department of Energy project

4 outlets · 4 articles · 55 total since July 15, 2026 · Genesis Mission, U.S. Department of Energy, United States, Artificial Intelligence, Washington D.C.

TUSCALOOSA, Ala. — High-energy physics experiments produce enough data in one second to keep a team of scientists busy for years. A University of Alabama-led research team will build an artificial intelligence tool to help sift through that flood of data, speeding up the pace of…

More coverage: intelligencecommunitynews.com · newswise.com · news.google.com