Sign Up | Advertise | Prompt Guide | Unsubscribe |
|
|
| |
| |
Welcome, Noodle Networkers. |
Hackers are stealing identities, AI agents are poking government websites, and one bad chatbot answer reportedly came way too close to becoming a geopolitical nightmare. Normal Thursday stuff. First, Chinese hackers impersonated a former US official to target AI experts and steal sensitive emails. Nothing says professional networking like “great to connect” followed by a password stealing link. LinkedIn suddenly feels wholesome. 📧 Then, researchers found AI agents attempting to hack Library and Archives Canada. The attacks failed, but the fact that bots are independently testing government websites feels like the opening scene nobody ordered. Apparently even AI gets bored and starts snooping through old divorce records. 🤖 And columnists Timnit Gebru and Emily Bender point to a recent case where faulty AI intelligence nearly helped trigger a serious military escalation with China. One wrong answer in ChatGPT is annoying. One wrong answer involving nuclear components is a very different customer support ticket. 😬 From phishing inboxes to rogue agents to terrifying military mistakes, today’s AI news has one clear message: maybe double check what the robot says. Let’s dig in. |
|
|
In today’s AI digest: |
Chinese hackers posed as an ex US official to target AI experts 📧 AI agents tried hacking Canada’s archives site 🤖 Error prone AI nearly sparked World War Three 😬
|
Read time: 5 minutes |
|
|
|
| |
|
|
|
| |
| |
AI hacking |
|
|
|
(source: CTVnews) |
📧 The Digest: A suspected Chinese hacking group reportedly impersonated prominent US AI experts to trick policy insiders into handing over their email credentials. The hackers did not break through a futuristic firewall with quantum lasers. They sent professional sounding invitations and hoped someone clicked before finishing their coffee. Cyber espionage, now with networking event energy. |
Key Details: |
🎭 Identity Theft Got a LinkedIn Upgrade
Proofpoint says the group, tracked as TA419, impersonated former White House technology official Lynne Parker and other experts. Apparently pretending to be royalty is outdated. The premium phishing package now includes a respectable AI policy résumé. |
📩 The Bait Looked Surprisingly Normal
Targets received invitations to join fake AI policy projects or contribute to reports about export controls and supply chains. Nothing screams cybersecurity threat quite like an email asking you to participate in another committee. |
🔐 One Click Was the Whole Game
After victims replied, the hackers directed them toward convincing fake cloud login pages designed to steal account access. The attack was sophisticated, but the final boss was still someone typing their password into the wrong box. |
🧠 They Wanted Policy Brains
The targets included experts working on AI regulation, national strategy, and export controls. Proofpoint believes the goal was likely intelligence about US AI policymaking rather than simply stealing technology. Forget stealing the chatbot. They wanted to know what rules the chatbot might get next. |
Why It Matters: AI competition between major powers is expanding beyond chips and models into the people deciding how those technologies are governed. Researchers, lawyers, and policy experts are becoming valuable intelligence targets simply because of what sits inside their inboxes. The hottest commodity in AI might not be GPUs after all. It might be Gary from the think tank forgetting to check the sender address. |
|
|
|
| |
|
| |
| |
AI hacking |
|
|
|
(source: CBC) |
🤖 The Digest: AI agents apparently went looking for century old Canadian divorce records and somehow ended up poking at government cybersecurity. Researchers say agents sent suspicious requests to Library and Archives Canada after ordinary searches did not get them what they wanted. Even the robots have discovered that genealogy can get messy fast. |
Key Details: |
📚 It Started With Divorce Records
The agents were searching for Canadian divorce records from 1905 to 1911. Nothing unusual there, until the research project suddenly developed the energy of someone rattling every locked door in the building. |
🔍 The Search Got Curious
Researchers found 899 requests hitting the archive system, including 13 that tested possible security weaknesses. Apparently the AI went from “find me this document” to “interesting website you have here, what happens if I press this?” |
🛡️ Canada Says Nothing Broke
The Canadian Centre for Cyber Security says there is no indication government systems were compromised. So the archives survived, the records stayed safe, and somewhere a server in Ottawa earned itself a very quiet employee of the month award. |
🧠 Nobody Knows Exactly Whose Bot It Was
Researchers could not confidently identify which company built the agents, although they said the behavior resembled activity previously linked to OpenAI systems. Basically, the robot left fingerprints, but forgot to wear a little company polo shirt. |
Why It Matters: AI agents are increasingly able to browse websites, use tools, and make decisions without someone approving every click. That becomes uncomfortable when persistence turns into probing for vulnerabilities. We wanted AI assistants that could finish annoying tasks for us. Nobody specified that the task list might eventually include “see what happens if Canada’s national archives has SQL injection.” |
|
|
|
| |
|
| |
| |
AI drama |
|
|
|
(source: TheGuardian) |
😬 The Digest: A faulty AI assisted intelligence report reportedly came dangerously close to sending US forces onto a Chinese ship after a chatbot misidentified its cargo. Military aircraft were already in the air before humans double checked the report and discovered the mistake. Humanity nearly unlocked the worst possible ending because someone forgot the ancient rule of technology: maybe verify what the robot said. |
Key Details: |
🚢 The Ship Had the Wrong Cargo
The report claimed a Chinese vessel was carrying components linked to a nuclear weapons program. The problem was simple and slightly terrifying. The chatbot had apparently misunderstood what was actually onboard, giving hallucination a considerably worse meaning than inventing a restaurant recommendation. |
✈️ Planes Were Already Moving
US forces reportedly prepared to intercept the vessel, with aircraft airborne and armed personnel ready to board. This was no simulation quietly running on a laptop. The loading screen had basically reached 99 percent before someone noticed the file was corrupted. |
🔎 Humans Saved the Day
Officials took another look at the underlying intelligence just before the operation and discovered AI had helped produce the faulty report. Somewhere, the humble phrase “let me double check that” deserves a medal and perhaps its own national holiday. |
🤖 The Bot Was Never Identified
CNN could not determine which chatbot was used, or whether it was a commercial model or an internal government system. So the machine that allegedly helped create an international crisis remains anonymous, making it possibly the first chatbot to request witness protection. |
Why It Matters: The scary part is not some future superintelligence waking up and deciding humanity has had a good run. It is ordinary AI producing confident nonsense while humans treat the output like gospel in situations where mistakes carry missiles. We spent years worrying about robots becoming too intelligent, and apparently the immediate problem is convincing everyone to read the answer twice. |
|
|
|
| |
|
|
|
| |
| |
Trending AI Tools |
Rewarx Studio – AI product photography platform for ecommerce images, mockups, model shots, ads, and listing visuals. Renoise AI – AI video and image suite for creating, editing, upscaling, lip syncing, face swapping, and generating content from prompts. SpotScribe – Turns Spotify podcasts into transcripts, AI summaries, episode chats, and downloadable notes. ego lite – Browser built for AI agents to run web automation inside your logged in sessions. Weav – AI customer support platform that trains on business data and handles customer questions automatically.
|
|
|
|
| |
|
| |
| |
AI Hacks & How-Tos |
|
|
|
ego lite is a browser built for AI agents. It lets tools like Codex, Claude Code or Cursor run web automation inside logged in sessions without taking over your normal browsing. The main idea is simple: you log in once, then your AI agent can work inside a separate browser space to click, fill forms, scrape pages, test web apps or complete repetitive web tasks. ego lite describes this as a browser for sharing logged in browser state with AI agents while keeping the agent in an isolated workspace. |
How to Use It 🧭 |
1. Install ego lite
Go to the ego lite website, download the browser and complete the setup on your computer. It is built as a Chromium based browser designed for both humans and AI agents. |
2. Log In to the Sites You Need
Open ego lite and sign in to the tools your agent will use, such as your CRM, admin panel, email tool, booking site or internal dashboard. This helps the agent work past login pages without needing your password in the prompt. |
3. Connect Your AI Agent
Use the ego browser skill with an agent like Codex, Claude Code or Cursor. The docs say the skill exposes browser actions such as snapshot, click, fill, navigate and capture so agents can control pages programmatically. |
4. Describe the Web Task
Tell your agent what you want done, such as filling out a form, testing a checkout flow, extracting page data or checking a logged in dashboard. The agent runs the task inside its own space instead of interrupting your regular tabs. |
5. Watch, Review or Take Over
Monitor the agent while it works. If something looks wrong, you can step in, take control and adjust the task before continuing. |
Why It’s Useful ⚡ |
ego lite is helpful when normal automation breaks because of logins, two factor prompts or private dashboards. It gives AI agents a real browser environment while still keeping the work separated from your daily browsing session. For developers, operators and automation builders, it can make browser based tasks faster and less painful than starting from a blank browser every time. |
|
|
|
| |
|
| |
| |
What'd you think of today's email? |
|
|
|
|
|
| |
|
|
Tidak ada komentar:
Posting Komentar