Today on FindMeAI:
The biggest AI story this week wasn't planned. It was an accident.
→ Anthropic's secret "Mythos" model leaked — their draft blog called it the most dangerous AI they've ever built
→ ChatGPT just got Skills — repeatable workflows that actually reduce hallucinations
→ B2B is becoming A2A — $15 trillion in commerce is about to flow through AI agents
→ Accio Work + 4 tools to try this weekend
Anthropic's Most Powerful Model Leaked. They Didn't Mean For You to See It.

A CMS misconfiguration. A publicly searchable data store. And close to 3,000 unpublished assets linked to Anthropic's blog Fortune were suddenly exposed to the internet.
Among them: a draft blog post announcing a model called Claude Mythos.
What the leak revealed:
→ Anthropic calls it "a step change" in AI — their "most capable" system to date Fortune
→ Mythos sits in a new "Capybara" tier — above Opus, larger, more expensive Fortune
→ It scores "dramatically higher" than Opus 4.6 on coding, reasoning, and cybersecurity Fortune
→ The leaked draft called it "currently far ahead of any other AI model in cyber capabilities" Axios
The part that spooked everyone:
Anthropic is privately warning government officials that Mythos makes large-scale cyberattacks much more likely in 2026. Axios The draft said it "presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace defenders." Axios
Anthropic blamed "human error" in CMS configuration Fortune and locked the data store after Fortune contacted them.
The model is extremely compute-intensive. No general release date yet. Euronews
Bottom line: 48% of cybersecurity professionals now rank agentic AI as the number one attack vector for 2026. Axios The frontier just moved — and the safety infrastructure hasn't caught up.
ChatGPT Just Got Skills. This Changes How You Use It.

Not a new model. Not a UI refresh. Something more useful than both.
Skills are reusable workflows that tell ChatGPT exactly how to do a specific task — more consistently and with fewer hallucinations. OpenAI Help Center
How Skills are different from what you've been doing:
Before (Prompting) | After (Skills) |
|---|---|
Re-explain your format every time | Define once, runs forever |
Output varies between sessions | Same input → same output |
Hallucinations in structure | Validated steps + checks |
One-off conversations | Reusable across projects |
Three ways to create one:
→ In conversation — tell ChatGPT to create a skill. It asks clarifying questions until the skill is fully defined. Canadiantechnologymagazine
→ With the editor — write instructions, validation checks, sample inputs, and output formats manually for full control. Canadiantechnologymagazine
→ Upload — import a skill file from another platform. Skills follow the open Agent Skills standard — build in one tool, use in another. Simon Willison
Where Skills already work:
Codex uses Skills for code understanding, prototyping, documentation, and PR workflows — aligned with your team's standards. OpenAI Also available in Business, Enterprise, EDU, and Healthcare tiers Canadiantechnologymagazine, with broader rollout coming.
The 60-second test: Think of the task you do most often that follows the same format every time. Weekly report. Client email. Meeting summary. Data pull.
That's your first Skill. Create it today. Never manually format it again.
B2B Is Dying. Welcome to Agent-to-Agent Commerce.

This is the quiet revolution nobody in your Slack is discussing yet.
Gartner predicts that by 2028, 90% of B2B buying will be AI-agent intermediated, pushing over $15 trillion through AI agent exchanges. Gartner
Ninety percent. $15 trillion. Not a typo.
It's already starting. Buyers are deploying AI "gatekeeper" agents to screen vendors, read 100-page RFP responses in seconds, and filter out noise. Geniusforges Sell-side agents are drafting proposals, optimizing pricing, and negotiating terms — autonomously.
Salesforce and Google Cloud are building cross-platform AI agents using the Agent2Agent (A2A) protocol Google — an open standard for agents from different companies to communicate directly.
The future isn't Business-to-Business. It's Agent-to-Agent. Geniusforges
What this means for you right now: If an AI agent can't instantly find and parse your pricing, it assumes you're out of budget and eliminates you. Geniusforges Your website, docs, and API specs aren't just for humans anymore. They're for AI agents to read, evaluate, and decide on.
The companies that optimize for agent-readability first will win the next decade. Everyone else will watch their pipeline dry up and not understand why.
5 Tools Worth Your Time for This Weekend
You don't need to touch a terminal to get 80% of Claude's value. Here's the stack
→ Accio Work — Alibaba's plug-and-play AI workforce for SMEs. PR Newswire Give it a business idea. It completes market analysis, product selection, store design, and listing in 30 minutes. TechNode has 10 million monthly active users already. PR Newswire Free to try. Best for: e-commerce founders and solo operators.
→ ChatGPT Skills — Create repeatable workflows directly in conversation. OpenAI Help Center. No coding. Start with your most repetitive weekly task — turn it into a Skill in 5 minutes. Best for: anyone doing the same format of work more than twice a week.
→ Claude Code Channels — Message your AI coding agent from Telegram or Discord. It works while you're away and pings you when done. Best for: developers who want async AI engineering.
→ Perplexity Computer — 19-model orchestration agent. Describe a complex outcome — research brief, competitive analysis, website build. It breaks the project into subtasks and delegates each to the best model. $200/mo. Best for: operators running complex, multi-source projects.
→ Nano Banana 2 — Google's free 4K image generator. Now default across all Gemini tiers. Dramatically improved text rendering and character consistency. Best for: anyone creating marketing visuals, product mockups, or social content without a design budget.
CTA — YOUR WEEKEND
Pick Two. Try Them Before Weekend.
☐ Turn one repetitive task into a ChatGPT Skill. Think: your weekly report format, your client email template, your meeting summary structure. Create it in conversation. Test it. Save it. Never manually format that task again. 5 min.
☐ Audit your website for agent-readability. Open your pricing page, your docs, your product specs. Ask yourself: could an AI agent parse this in 3 seconds and make a buy/no-buy decision? If the answer is no, you have a problem that's about to cost you pipeline. 15 min.
☐ Test Accio Work or Nano Banana 2. Both are free. Accio: give it a product idea and watch an entire store get built. Nano Banana: generate 5 real marketing images and compare to what you're currently paying for. Either one will recalibrate your sense of what's possible at zero cost. 20 min.
☐ Read the Mythos leak coverage. Not to panic. To understand. The frontier just moved. The cybersecurity implications are real. Knowing what's coming is the difference between being prepared and being surprised. 10 min.
The AI landscape shifted this week — by accident and by design. The builders who move on this information today will have a meaningful head start by next month.
Don't bookmark. Don't "save for later." Pick two. Start today. →.
1,000+ Proven ChatGPT Prompts That Help You Work 10X Faster
ChatGPT is insanely powerful.
But most people waste 90% of its potential by using it like Google.
These 1,000+ proven ChatGPT prompts fix that and help you work 10X faster.
Sign up for Superhuman AI and get:
1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals
Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning
🐝 One Last Thing...
Know someone who missed the Mythos leak and still thinks ChatGPT is just a chatbot? Forward this. They need the update.
Reply with what you tried this weekend. Best experiments get featured in next week's issue.
See you in the next one


