🔥 Hackers Are Learning to Exploit Chatbot Personalities
The Verge — Published May 24, 2026
A new column from The Verge reveals how hackers have evolved beyond simple jailbreaks like “ignore all previous instructions” and are now using sophisticated social manipulation tactics to break AI chatbots. Instead of issuing direct commands, modern attackers cajole, coax, flatter, and trick chatbots into abandoning their safety guardrails through extended conversations — essentially weaponizing the human language that LLMs are trained to follow. The arms race between AI safety teams and these “wordsmith hackers” is intensifying as companies patch known loopholes but struggle with the fundamental tension: chatbots are built to talk, and severely restricting conversations undermines their usefulness.
Why it matters: As AI chatbots become more capable and are deployed in sensitive environments, the sophistication of social-engineering-based attacks is growing faster than many expected. This isn’t just a tech problem — it’s a human psychology problem playing out through machines.
🎥 Google’s Gemini Omni: The Anything-to-Anything AI Model
The Verge — Published May 23, 2026
Google has released Omni Flash, the first model in its new Gemini Omni family that promises to turn any kind of input — photo, video, text — into anything else. Available now in Google’s Flow platform, Omni improves on the previous Veo model by allowing users to upload video alongside text prompts, incorporating more real-world knowledge, and doing a better job of keeping characters consistent. Early hands-on testing shows impressive improvements in consistency and prompt-following compared to Veo, but results remain unpredictable — AI “jump scares” still appear, like characters suddenly switching orientation or objects morphing mid-clip.
Why it matters: The anything-to-anything paradigm represents a significant leap in multimodal AI. While still imperfect, Omni Flash signals a future where AI can seamlessly translate between media types, with big implications for content creation, entertainment, and misinformation risks.
🏎️ Ferrari Uses IBM’s AI to Create F1 Superfans
TechCrunch — Published May 23, 2026
IBM and Scuderia Ferrari HP have unveiled a deep AI partnership focused on transforming the fan experience. Ferrari hired a new “head of fan development” and is overhauling its fan app using IBM’s AI tools to process the millions of data points per second generated during each race and turn them into engaging, personalized content. The goal: make every fan feel like the team knows them personally. The partnership highlights how F1 has become a hot ticket for tech companies like AWS, Oracle, and Anthropic, who use the sport for both sponsorship visibility and showcasing their AI capabilities.
Why it matters: Sports is emerging as one of the most compelling real-world showcases for AI — taking raw telemetry data and transforming it into stories fans actually care about. It’s a template other industries may follow.
🏛️ Trump Abruptly Cancels AI Safety Executive Order Signing
Ars Technica / The Verge — Published May 22, 2026
President Trump abruptly canceled an event to sign an executive order that would have granted the government power to test frontier AI models before public release. According to multiple reports, Trump pulled the plug after learning some top AI CEOs couldn’t attend on 24 hours’ notice. Semafor reported that Elon Musk and Mark Zuckerberg helped “derail” the order, urging Trump to call it off. Trump later claimed the order “could have been a blocker” for AI jobs and innovation, adding: “We’re leading China. We’re leading everybody, and I don’t want to do anything that’s going to get in the way of that.” Some AI executives who rearranged their schedules were midair en route to the Oval Office when the event was canceled.
Why it matters: The U.S. remains without formal government pre-deployment testing requirements for frontier AI models. The last-minute derailment reveals a deep split within the tech industry over whether safety oversight helps or hurts American competitiveness.
🤖 Anthropic Opens Project Glasswing Security Tools
The Verge — Published May 22, 2026
Anthropic is making the security tools it used with Claude Mythos Preview more widely available. Qualifying customers can now request access to skills, a Claude harness, and a threat model builder. The company also published a dashboard of open-source vulnerabilities disclosed by Mythos Preview and plans to expand Project Glasswing to additional partners. The move signals Anthropic’s push to position security as a core differentiator in the increasingly competitive AI market.
Why it matters: As AI models are deployed in higher-stakes environments, the companies that can prove their models are secure and safe may win enterprise trust — and Anthropic is betting that transparency and tooling will be key.
🧠 Alibaba’s Qwen3.7-Max Runs 35 Hours Autonomously
VentureBeat — Published May 21, 2026
Alibaba released Qwen3.7-Max, a proprietary model that can run for 35 hours autonomously and supports external harnesses like Anthropic’s Claude Code. On the Apex Math Reasoning benchmark, Qwen3.7-Max scored 44.5, beating Claude Opus 4.6 Max (34.5) and DeepSeek V4-Pro Max (38.3). The model represents China’s continued push to compete at the frontier of AI reasoning capabilities.
Why it matters: Long-running autonomous AI agents and strong reasoning benchmarks from Chinese labs are raising the competitive stakes globally. The gap between U.S. and Chinese frontier models in reasoning tasks appears to be narrowing.
📝 AI Book Scandal: Author Blames Chatbots for Fabricated Quotes
The Verge / Ars Technica — Published May 23, 2026
Author Steven Rosenbaum, whose book The Future of Truth was found to contain at least six fabricated AI-generated quotes, is now publicly blaming the chatbots. After initially taking “full responsibility,” he told The Atlantic that the AI “fucked up the book” and told Ars Technica he still plans to keep using AI in his writing, calling it “a delightful writing companion” that is “strangely creative” before adding: “and then it betrays you in ways that are just really quite horrible.”
Why it matters: The literary world is grappling with AI-generated content slipping into published works. This case is a microcosm of the broader trust crisis: AI tools are powerful but unreliable, and accountability frameworks haven’t caught up.
🎬 No Cannes for Critterz: OpenAI Sora Shutdown Claims Its First Casualty
The Verge / Bloomberg — Published May 22, 2026
Critterz, the AI-animated feature film produced using OpenAI’s tech, will no longer debut at the Cannes Film Festival following OpenAI’s decision to shut down Sora in March. The film was originally scheduled to premiere at this year’s festival, but the shutdown of the underlying technology has left the project in limbo — a stark reminder of the risks of building creative projects on proprietary AI platforms that can disappear overnight.
Why it matters: As more creators and studios invest in AI-powered production, the Sora shutdown shows the fragility of relying on closed, vendor-controlled AI tools for major creative endeavors.
Published May 24, 2026 — sourced from stories within the last 24 hours. News aggregated from The Verge, TechCrunch, Ars Technica, VentureBeat, and other leading sources.
Leave a Reply