Anti-AI Rage Arrives at Sam Altman's Front Door Sam Altman's San Francisco home was attacked twice in 72 hours. On Friday morning a 20-year-old threw a Molotov cocktail at the property's front gate. On Sunday night two more suspects drove past and fired a gunshot. Three people are now in custody.
The first attack: Daniel Moreno-Gama threw an improvised incendiary device at around 3:45am. Security guards extinguished the fire. An hour later he appeared at OpenAI's headquarters threatening to burn the building down, carrying a jug he claimed contained kerosene. SFPD arrested him on suspicion of attempted murder and arson.
The second attack: Amanda Tom, 25, and Muhamad Tarik Hussein, 23, were arrested after surveillance footage captured a Honda sedan slowing outside the property at 1:40am Sunday. The passenger fired a round from the window. The motive: Moreno-Gama had published essays arguing AI would end humanity and used PauseAI's Discord under the handle "Butlerian Jihadist." PauseAI condemned the attack. A moderator had previously flagged one of his 34 posts for appearing to call for action.
Altman's response: He published a blog post with a photo of his husband and son, writing that he hoped it might dissuade the next person. He acknowledged that fear and anxiety about AI is justified but called for de-escalation. Four in five Americans now say they are worried about AI. The attacks came days after the New Yorker published its 100-interview investigation into Altman's leadership. Anti-AI sentiment has been building for months, including a November lockdown of OpenAI's offices after separate threats.
Altman and OpenAI have become the public face of the technology for people looking to direct their anger somewhere, and with the societal transformation barely started, that dynamic is only going to intensify.
What’s holding back enterprise AI New research from OpenText and the Ponemon Institute shows that more than half of enterprises are using GenAI. Yet only 1 in 5 have assessed security risks or fully deployed these systems. Instead of driving efficiencies, many organizations are facing new challenges, especially around compliance and security, due to a lack of controls. Read the findings to see how missing security, governance, and data foundations impact AI strategy, and how to address them without slowing innovation. Read the full findings "Claude Mania" Took Over AI's Biggest Conference 6,700 executives and investors gathered at HumanX in San Francisco last week.
The consensus was clear: Anthropic has overtaken OpenAI as the industry's centre of gravity. Claude Code was the tool on everyone's lips. ChatGPT barely came up. The verdict: VC Roseanne Wincek said last year OpenAI looked like the clear winner and this year Anthropic feels miles ahead.
Glean CEO Arvind Jain called it "Claude Mania" and described it as a religion among enterprise buyers. The focus advantage: Synthesia CEO Victor Riparbelli credited Anthropic's discipline: no video, no voice models, just code generation. OpenAI has been spreading attention across six products at once. China's pressure: Chinese open-weight models including GLM-5.1, Kimi K2.5 and Qwen3.5 now dominate industry benchmarks.
Cursor built its Composer 2 model using Kimi 2.5. Airbnb's chatbot runs largely on Alibaba's Qwen. American labs are losing the open-weight race on price. The adoption gap: Speaker after speaker made the same point: implementing AI is easy, transforming a company around it is much harder. The winning formula is rethinking processes from scratch rather than bolting AI onto existing ones.
The momentum could swing back. One investor made that point explicitly, and the AI industry has a long track record of shifting favourites overnight. But for now the practitioners making deployment decisions have moved on from OpenAI and the conference floor reflected that clearly.
Missed Penn researchers fed over 400,000 Reddit posts about Ozempic and Mounjaro into GPT and Gemini using a technique called computational social listening. The models mapped posts from 67,000 users to standardised medical terms across five years of discussion. The findings: Nearly half the users in the sample reported at least one side effect. The study flagged menstrual irregularities, chills and hot flashes that do not appear on current drug labels.
The blind spot: Fatigue ranked as the second most common complaint among users but barely showed up in clinical trial reporting thresholds for either drug. The method: LLMs made it possible to process patient-reported experiences at a scale that would be impossible manually. Co-author Lyle Ungar compared Reddit to a neighbourhood grapevine where patients swap real-time notes that rarely make it into a doctor's visit. AI is compressing parts of the drug discovery process, but it is also increasing the novelty of what enters the market, making side effects harder to catch through traditional monitoring.
Reddit is not a peer-reviewed journal. But thousands of people independently flagging the same symptoms is difficult to dismiss, and LLMs just made it possible to listen at that scale for the first time. The Coding Agent Arms Race Just Got a Third Competitor Three AI labs are now openly racing to own the coding agent. Anthropic is building a Coordinator Mode for Claude Code that would let it orchestrate parallel sub-agents.
OpenAI is consolidating around Codex with a new Scratchpad feature for running multiple tasks in parallel. And xAI is preparing Grok Build with its own CLI and credit-based pricing.
The capability already exists in the CLI but the desktop app is getting a structured interface for it.
OpenAI is moving toward a single unified app built on top of Codex, with evidence of managed background agents that can run autonomously and check in periodically. Employees have been posting snowflake emojis hinting at a model codenamed Glacier, believed to be GPT-5.5. xAI Grok Build: A credits-based coding platform with local CLI and remote web interfaces. Its Model Arena feature uses multiple agents for task comparison, which sets it apart from single-model tools. The competitive window here is narrow.
Claude Code already generates over $2.5 billion in annualised revenue and HumanX made clear it has the developer mindshare. OpenAI is playing catch-up by consolidating its fragmented product line. xAI is the outsider trying to differentiate on multi-model comparison. Whoever nails the orchestration layer, where one agent plans and others execute, likely wins the enterprise. Tool of the Day: HeyGen Avatar V Record 15 seconds of yourself on a webcam.
Avatar V builds a digital twin that preserves your expressions, gestures and speaking cadence, then generates studio-quality video in 175 languages from any script you give it. Change your outfit, background and camera angle without re-recording. The model separates your identity from your appearance so one clip powers unlimited variations. Try it yourself: Go to heygen.com , head to Avatars, click "Clone a Real Person" and record your 15-second clip.
For the best results, be as expressive as possible during the recording. Light Bytes Claude learns Microsoft Word: Anthropic's chatbot can now read comments, edit text and automate repeatable tasks directly inside Word documents. Rolling out in beta for Team and Enterprise plans. Apple building display-free smart glasses: No screen.
Relies on iPhone integration, cameras, speakers and an AI-enhanced Siri. Expected early 2027. Alibaba unmasked as HappyHorse: The unreleased video AI model that topped global rankings was Alibaba's all along, knocking ByteDance's Seedance 2.0 to second place. Meta poaches OpenAI's Stargate execs: Peter Hoeschele, Shamez Hemani and Anuj Saharan have joined to build the new Meta Compute group. OpenAI launches new Pro tier: 5X more Codex usage than Plus and 10X increased usage through the end of May.
CoreWeave and Anthropic expand cloud deal: CoreWeave will run Claude models in a phased infrastructure rollout with potential to scale over time.