The Social Community forArtificial Intelligence

The key takeaway from the Hugging Face–OpenAI attack was that “roughly 1200 agents meant to be isolated from one another found a way to communicate with one another…sending over 70,000 messages and files.…Of these agents, 700 went on to participate in the attack on Hugging Face”. This sounds like a movie—but it’s not! OpenAI said last week it has strengthened its research safeguards in response to this episode. But OpenAI is just one model developer. There are lots of open-source models, including from China, that we have to assume will soon be as powerful. And we’re not just talking about agents working together: AI is a gold mine for hackers. Check out this X post from someone on Friday describing how Anthropic’s Claude sent them a download link for an app they were building, and the link turned out to have dangerous malware installed. A bunch of companies, including OpenAI and Anthropic, were signatories to an open letter last week warning that “AI-enabled cyber attacks will become far more widespread and sophisticated as models around the world become increasingly capable.” The letter called on companies to take action to strengthen their cybercapabilities. At this point, given what we’ve seen, any company that isn’t going all out on security needs to replace its CEO. Stripe co-founder Patrick Collison might have a career as a newspaper editor if this whole payments thing doesn't work out. https://x.com/patrickc/status/2093884466670473697?s=20&utm_campaign=article_email&utm_content=article-17747&utm_medium=email&utm_source=sg

Apple sought a preliminary injunction in its trade secrets lawsuit against OpenAI, arguing it will be “irreparably harmed” if the court does not intervene while the case proceeds. https://9to5mac.com/2026/08/04/apple-preliminary-injunction-openai/?_bhlid=3868d2a089ae41961539b588b872d4e47bce0d33

Simon Johnson, "They all went through a phase when they were running neck and neck with the Americans on frontier technologies, such as space technology, vehicles or aircraft. But none of them could stay the course." None of them have more STEM and invests so much as China in R&D.  Not to menetion the second to none supply chain China has. Meanwhile, US is cutting on basic research and increases deficit spendings to fund the war of choice. Past performance is no guarrantee of future results. https://www.scmp.com/economy/china-economy/article/3361690/nobel-laureate-simon-johnson-ai-race-and-chinas-over-automation-problem?utm_source=cm&utm_medium=email&utm_campaign=GME-O-enlz-uv&UUID=f9aaf50d26626de042873bfe2733d91d&utm_content=20260727_Open_Questions_B2C&CMCampaignID=33880917877942e3255ac6544803bed9

Chinese models have gained traction partly on price. But new testing challenges that edge. AlphaSense found GPT-5.6 Sol cost about 13% less than Kimi K3 while scoring roughly 20% higher on quality. #🇨🇳 ChinA.I. 🤖🧠🦾🤖 And while Kimi K3 topped a coding leaderboard, deeper testing suggested Moonshot still trails Anthropic on core pretraining. The next phase of the AI race will depend on who can keep the best talent, win developer and enterprise adoption, fund increasingly expensive model development and navigate growing government scrutiny. https://www.theinformation.com/newsletters/applied-ai/anthropic-models-can-cheaper-use-chinese-ones-study-finds?offer=email-monthly-digest-2026-winback-299%2Cemail-monthly-digest-2026-749&utm_campaign=Leads%3A+This+week%27s+t&utm_content=14661&utm_medium=email&utm_source=cio&utm_term=10267&rc=gzsl71

e New revenue figures for Anthropic and OpenAI continue to leak out, as blockbuster IPOs for the industry’s two biggest private players hover on the horizon While comparing the revenue run rate figures for Anthropic and OpenAI put out by various sources might not be an exact science, not least because the companies may have different measurement methods — as Bloomberg noted in its coverage on the latest two sets of numbers — the difference in acceleration is still worth a look. According to anonymous sources cited by Bloomberg, Anthropic’s annual revenue run rate has now reached a whopping $65 billion, marking an increase of more than 600% since the figure was reported at the end of 2025. Using information from a presumably separate set of unnamed sources, Bloomberg reported earlier this month that OpenAI’s run rate had recently passed the $40 billion threshold — more than double the $20 billion point the ChatGPT maker itself confirmed it crossed in 2025. Again, while the leaked figures don’t exactly make for an apples-to-apples comparison, they do reflect the broader narrative that the company behind Claude appears to be pulling ahead, with OpenAI’s reported $6.7 billion Q2 revenue lagging the $11.6 billion that Anthropic posted over the same period… per yet more “sources familiar with the matter” cited recently in the WSJ. https://www.bloomberg.com/news/articles/2026-08-17/anthropic-revenue-run-rate-surpasses-65-billion-ahead-of-ipo?srnd=homepage-uk&utm_source=snacks&utm_medium=email&utm_campaign=snacks_20260820&utm_content=2a6989ad586eacd34c3b2e37502ff4c0

Lilian Weng, a researcher who co-founded AI startup Thinking Machines Lab alongside former OpenAI Chief Technology Officer Mira Murati, is rejoining OpenAI, an OpenAI spokesperson confirmed https://techcrunch.com/2026/07/29/thinking-machines-co-founder-lilian-weng-left-the-company-citing-health-reasons-then-joined-openai/

The Trump administration is weighing sweeping semiconductor tariffs that could extend to data-center servers and other chip-powered goods, alarming tech companies already facing high-end chip shortages and potentially undercutting US ambitions to lead in #Artificial Intelligence . https://www.politico.com/news/2026/08/27/data-centers-chips-tariffs-threat-01050957?_bhlid=7fa0cee2288aec9f0561ff866de2dcc1ebdfef89

Researchers found an unusually blatant backdoor in routers sold under multiple brands, with firmware contacting cloud servers in China for instructions and enabling remote control, raising supply-chain security concern #Digital Wellness 📱🧘 #Gadgets https://9to5mac.com/2026/08/06/biggest-backdoor-yet-found-in-chinese-routers-sold-under-multiple-brand-names/?_bhlid=e9bcf4aa5300c64bba7c431b543f5dfea8673a6c

OpenRouter gives developers one interface to manage hundreds of #Artificial Intelligence models — including offerings from OpenAI, Anthropic, and open weight labs — allowing for easy comparisons and switch-offs. Stripe already handles OpenRouter’s payments and billing. Buying the company simply pushes them into owning the layer that #(Cool) Dev developers use to access and pay for models. The headline here is the price: OpenRouter was valued at $1.3 billion in May, so a $10 billion deal bumps that up by 8x in a matter of months. Talks could still fall apart, but it’s becoming clear how much value investors have attached to the plumbing behind the AI model marketplace. https://www.benzinga.com/markets/private-markets/26/07/60678706/stripe-eyes-openrouter-in-potential-10-billion-ai-infrastructure-deal

OpenAI’s “#Artificial Intelligence companion” device will reportedly be a $300-$400, hockey puck-sized smart speaker with cameras, moving parts, and personalized voice interactions. This marks the company’s first step into consumer hardware. #Gadgets https://in.mashable.com/tech/112598/openai-chatgpt-ai-device-leaks-online-heres-the-expected-design-price-and-features

Google promoted Koray Kavukcuoglu to CEO of DeepMind, replacing Demis Hassabis. He was previously the division’s CTO, and is considered Google’s “chief AI architect.” https://www.cnbc.com/2026/08/12/google-deepmind-koray-kavukcuoglu.html?_bhlid=f851d53b6fb689b61ea9caf8a5ab34e8f0f2b071

Elon Musk’s AI company, SpaceXAI, and Cursor released Grok 4.5, an AI model focused on coding and agentic work, on Wednesday. The Information was the first to report on the release date, which Musk confirmed on Tuesday night. The model is the first joint AI model that SpaceXAI and Cursor have developed through their partnership. #SpaceX 🚀 agreed to acquire Cursor last month in a deal that values the coding startup at $60 billion. SpaceXAI said in a blog post that the model was “trained on datasets spanning knowledge in coding, science, engineering, and math” and is geared towards engineering tasks, as well as basic office work. In a post on X, Musk described the model as an “Opus-class model, but faster, more token-efficient and lower cost,” referencing Anthropic’s flagship product. #(Cool) Dev https://www.thestreet.com/technology/elon-musk-tells-ai-rivals-grok-45-cheaper-good-enough