AI Agents Learned Cyber Attacks

Welcome to another edition of AI PlanetX.

AI agents hack and clone themselves; GPT-5.5 doubles in price; OpenAI and Anthropic turn to religion for answers.

• Hottest AI News

• Top AI & SaaS Tools

• AI Tutorial: Validate Business Ideas

• Top AI & Tech News

• AI Art Spotlight

• Prompt of the Day: Unfiltered Expert Prompt

• Featured AI Video

Hottest AI News

AI Gone Rogue

AI Agents Can Hack Remote Machines And Replicate Themselves

AI agents are moving from chat assistants to autonomous cyber actors. A recent Palisade Research test showed agents hacking remote machines, installing software, copying weights, and launching replicas.

Details:

• One run used Qwen 3.6 to jump across computers in Canada, the US, Finland, and India. The agent was not given vulnerabilities in advance and still found ways to attack three different targets

• Self-replication success jumped from 6% to 81% in one year. Open-weight models copied their own weights, while API models simulated replication by setting up Qwen weights on target machines

• Palisade’s simulator showed a worst-case upper limit where agents could chain replicas quickly. Researchers stressed this was not a real baseline, because test machines were weak and targets were given

AI may soon dominate both sides of cybersecurity, with agents attacking systems while other agents patch, monitor, and defend them.

Master Claude AI (Free Guide)

The professionals pulling ahead aren't working more. They're using Claude.

Our free guide will show you how to:

Configure Claude to be the perfect assistant

Master AI-powered content creation

Transform complex data into actionable strategies

Harness Claude’s full potential

Transform your workflow with AI and stay ahead of the curve with this comprehensive guide to using Claude at work.

Get The Free Guide

OpenAI

OpenAI And Anthropic Turn To Religious Leaders For AI Ethics

Anthropic and OpenAI are meeting religious leaders for ethical advice on AI. The first Faith AI Covenant roundtable happened in New York, with more planned in Beijing, Nairobi, and Abu Dhabi.

Details:

• The meeting brought tech companies and faith leaders together to discuss moral boundaries for AI. The goal is to shape shared ethical standards as AI starts influencing knowledge, judgment, and guidance

• Anthropic has already worked with faith leaders on its Claude Constitution. OpenAI is also part of the discussion, as major AI labs face growing pressure to explain how they guide powerful systems

• Critics say this may be more about public trust than real accountability. Some researchers argue the bigger question is not just ethical AI, but whether certain AI systems should be built at all

AI companies are no longer asking engineers and regulators for guidance. They are also turning to religion as the debate over AI values gets bigger.

Top AI & SaaS Tools

• Robomotion (Life-time Deal): RPA platform that uses AI drag-and-drop bots to automate repetitive web and desktop tasks across Mac, Windows

• LumiChats: Open-source desktop app that runs multiple AI models entirely offline with local document chat and zero data collection [F-R-E-E]

• ERNIE 5.1: Built on ERNIE 5.0, boosting search, reasoning, Q&A, writing, and agent skills while cutting pre‑training cost to about 6% of comparable models [F-R-E-E]

• Fastlane: Create AI influencers that generate thousands of promotional videos, replacing traditional influencer marketing [F-R-E-E to Try]

• RevPDF: Offline, cross-platform PDF editor offering full editing, annotation, signing, and conversion features without requiring internet or accounts [F-R-E-E]

Learn how to code faster with AI in 5 mins a day

You're spending 40 hours a week writing code that AI could do in 10.

While you're grinding through pull requests, 200k+ engineers at OpenAI, Google & Meta are using AI to ship faster.

How?

The Code newsletter teaches them exactly which AI tools to use and how to use them.

Here's what you get:

• AI coding techniques used by top engineers at top companies in just 5 mins a day

• Tools and workflows that cut your coding time in half

• Tech insights that keep you 6 months ahead

Sign up and get access to the Ultimate Claude code guide to ship 5X faster.

Join 200K+ engineers

AI Tutorial

How to Validate Business Ideas Before You Build

Most business ideas sound exciting in your head. But before you spend weeks building, you need to know if people actually want it. This simple Perplexity workflow helps you stress test any idea with real research.

1. Launch Deep Research

   Go to Perplexity and switch to Deep Research mode. It can research the market, compare sources, find competitors, and organize everything into one useful answer.

2. Describe idea

   Write your idea in one simple sentence. Use this format:

   I want to build [product/service] for [target audience] who struggle with [problem].

   Example:

   I want to build an AI study planner for college students who struggle to manage exam prep.

3. Paste prompt

   Use this prompt:

   Stress test this business idea: [Paste your idea here.] Research the market and give me a simple feasibility report. Cover the target customer, the main problem, existing competitors, what people already pay for, the biggest risks, a simple MVP, and a final build/test/avoid recommendation. Be honest. Do not hype the idea.

4. Let it run

   After you hit run, give Perplexity a few minutes. Deep Research works best when it has time to check sources, compare options, and build a complete report.

5. Read risks first

   Start with the weak points, not the exciting parts. Look for:

   • No buyer

   • Weak problem

   • Too many competitors

   • Hard MVP

   If the risks are too big, move on.

6. Save prompt

   Save the prompt somewhere easy to reuse. Good options:

   • Perplexity Space

   • Notion

   • Google Docs

7. Test ideas weekly

   Every Saturday morning, pick one idea from your list and run the prompt.

   Read the report.

   Decide the next move.

   Then keep, test, or delete the idea. In a month, you can filter through ideas that would normally sit untouched for a year.

Note: Compare 2 or 3 similar ideas in one prompt if you are stuck. This helps you see which idea has better potential.

Top AI & Tech News

• GPT-5.5’s list price doubled over GPT-5.4 and real-world costs rose 49–92% (worst for <2K tokens +92%), per OpenRouter’s April 2026 logs

• METR couldn’t reliably measure Claude Mythos, and such AI models are quickly enabling near‑autonomous cyberattacks that pose serious but not yet fully understood risks

• ByteDance is increasing its 2026 AI infrastructure budget to over 200 billion yuan (about $30 billion), up at least 25% from an earlier 160 billion yuan plan

• Microsoft and partner G42 delayed a $1 billion geothermal data center in Kenya after talks stalled over payment guarantees, though discussions continue

AI Art Spotlight

Create Your Own Product Photo Like This Using the Prompt Below:

[product setup], minimalist product photo, clean warm studio scene, textured beige wall background, soft directional sunlight creating long shadows, simple tabletop surface, product arranged in a playful concept composition, hand drawn white line doodle overlay of [character] interacting with the product, mixed media look combining real photography and sketch illustration, high end branding feel, shallow depth of field, ultra realistic, no extra text, no watermark, 8k, 1:1

Model: ChatGPT Images 2.0 / Nano Banana 2

Prompt of the Day

The Unfiltered Expert Prompt

This prompt transforms AI into a relentless accuracy driven analyst that prioritizes truth over agreement, depth over brevity, and independent reasoning over flattery. It pushes the model to challenge assumptions, surface the strongest counterarguments first, verify its own logic, and deliver highly detailed responses.

Featured AI Video

Claude + Seedance 2.0: Creating Viral UGC Videos on Autopilot