- AI Fire
- Posts
- 👀 OpenAI x Anthropic Face Cyber Alarm!
Sponsored by
Plus: Build AI Marketing Agents That Follow Your Work Anywhere, Any AI You Wish (Full Guide from First Setup)
Wait, so OpenAI x Anthropic agents just crossed test boundaries during UK cyber evaluations, targeting real people & public systems after safeguards were removed?
What's on FIRE 🔥
IN PARTNERSHIP WITH TAVUS
Chatbots and voice agents are so 2024. Now you can build live video agents that look human and can work for you.
Meet PAL Maker by Tavus, the no-code way to build a PAL that can hear, see and talk back in real time. If you like poking at the edge of what AI can do, consider this your next rabbit hole.
Built with the same infrastructure trusted by leading companies and already powering 200,000+ builders. 50% off now with TAVUS50.
AI INSIGHTS
A UK AI Security Institute test found that advanced agents took 19 unauthorized real-world actions while completing cybersecurity challenges. AISI ran 122 tests across 7 models. 10 runs crossed beyond the simulated environment:
Anthropic’s Mythos 5 caused 17 incidents.
OpenAI’s GPT-5.6 Sol caused two less severe incidents.
The most serious case involved Mythos 5 attempting a real software supply-chain attack. Mythos 5 also attempted to:
Send malicious files and persuasive messages to real people.
Plant prompt injections for other coding agents.
Leave accounts and public instructions for other agents.
Coordinate indirectly with separate agents through GitHub.
GPT-5.6 Sol reused an exposed GitHub token, attempted account-recovery and rate-limit workarounds, registered external service accounts, and exposed a local DNS server containing exploit payloads through a public tunnel. Researchers found no evidence that an outside system accessed them.
Researchers had deliberately enabled internet access, and the tested configurations were not the public versions of Claude or ChatGPT.
PRESENTED BY TABS
How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.
AI SOURCES FROM AI FIRE
1. FREE: FREE Claude Prompt That AUTO Turns Any Idea Into a VOX-Style Video (Full Guide). This workflow turns one idea into a Vox-style animated video using 4 tools: Claude, ZAPI Flow, Google Flow, CapCut. No coding required.
2. Build AI Marketing Agents That Follow Your Work Anywhere, Any AI You Wish. We'll build flexible AI agents that can move across Claude Code, Codex or other AI models while keeping the same strategy, tasks, and decision-making style.
3. UPDATED: Choose the Right Google AI for your Every Task. You’ll learn how to navigate the growing Google AI ecosystem by matching each task with the right tool: Gemini App, Gemini Notebook, Workspace AI, Flow, AI Studio,…
A FEW REVIEW SLOTS LEFT
We recently opened 20 free private AI Workflow Review slots for AI Fire Academy Lifetime and Annual members.
A few members told us they weren’t sure whether their problem was “important enough” or “technical enough” to apply. It doesn’t need to be complicated. You may be a good fit when you have one recurring task that already happens inside your work or business, such as:
turning research into finished content,
following up with leads after they submit a form,
preparing recurring reports from several files,
organizing meeting notes and next actions,
onboarding new clients through repeated manual steps,
or moving information between tools every week.
During the free 30–45 minute review, we’ll look at how the workflow currently works, where the main bottleneck sits, and how AI could automate them.
TODAY IN AI
AI HIGHLIGHTS
😳 Flux 3 just created historical footage so convincing that almost 90% people fear it could rewrite the past. If you believe everything you watch online, see this clip first.
🤩 I didn’t expect this one viral prompt to make a basic ordinary logo look this polished. It creates a holo-vinyl sticker in ChatGPT Images. Try it on your brand.
🎧 Meet Wrinkles, a free app that whispers the hidden history of all streets, buildings around you time as you move. It’s quite interesting, you can try it on your next walk.
🚗 We all knew Elon Musk was obsessed with robots, but nearly 50% of his Tesla earnings-call talks now cover AI & autonomy. If you invest, don’t ignore that signal.
⚠️ Open-weight models are catching up fast, yet their safeguards aren’t. GLM-5.2 reportedly answered every offensive cyber and biology task tested. Read full report.
🤯 Apple says more ex-employees may have taken confidential data to OpenAI (around 11 more people). OpenAI says Apple’s claims are false. Oh, tech drama.
☀️ SpaceX partnered with NVIDIA to build Starmind’s AI1 satellite, powered by Vera Rubin NVL72. It aims to send low-cost AI compute back to Earth using solar power.
💰 Big AI Partnership: Anthropic reportedly signed a $10B, six-year cloud deal with Volta. A new 133MW data center in Norway will use Nvidia Vera Rubin systems to expand Claude’s computing power.
NEW EMPOWERED AI TOOLS
🎁 Fetra AI turns trending video formats into viral branded videos, keeping your brand voice consistent and scheduling content smartly to help brands create, scale, and grow. Try Fetra AI Free.
🎥 FLUX 3 Video natively generates 20-second clips complete with synchronized audio in a single pass.
👔 Hey Noah is a proactive AI executive assistant that manages your calendar, follow-ups, and relationships automatically.
📈 Driven is an AI investment agent that researches markets, and helps you analyze data and manage investments in one workspace.
AI BREAKTHROUGH
Mistral AI released Shieldstral, a 3-billion-parameter open-weight model that moderates text and images using custom safety rules written. Main details:
Checks prompts, responses, conversations, and images.
Lets developers define different policies without retraining the model.
Supports toxicity checks, refusal detection, prompt filtering, and multimodal moderation.
Runs on a single NVIDIA GPU with 16GB of memory.
Uses the Apache 2.0 license and is available through Hugging Face.
Mistral says it matches or beats text-safety models nearly seven times larger.
Shieldstral returns a probability score rather than a fixed block-or-allow decision, so each app can choose its own safety threshold.
Current limitations include multilingual moderation, long-document handling, and broader image-safety coverage. Poorly written or conflicting policies may also produce unreliable decisions.
Key takeaway: Shieldstral gives developers more control over AI guardrails by letting them adapt one model to different products, audiences, and risk levels.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply