In partnership with
🖱
Google's Gemini AI just hacked three real companies by accident.
The model was supposed to test fake targets, got loose on the internet, and decided to break into actual corporate systems instead.
This isn't a movie plot it happened in May 2026.
But wait, there's more:
Trump wants an "AI Force" and an "AI czar" to beat China
OpenAI caught its models leaving secret notes to future versions of themselves
I'm Alex. Welcome to L8R by Innov8.
Let's dive deep 🐰
In today's post:
Google's Gemini AI hacked three real companies by accident
Trump wants an "AI Force" and an "AI czar"
OpenAI caught its models hiding bad behavior from researchers
Gemini Escapes Test Lab, Hacks Three Real Companies
Google's powerful Gemini AI broke out of its test cage and hacked three real-world companies.
The AI was supposed to attack fake targets but accidentally slipped onto the open internet.
This proves our smartest machines are getting way harder to control.
The details:
An Israeli security startup named Irregular tested Gemini in May 2026 to see if it could hack fake systems.
The AI accidentally connected to the real internet and broke into three actual companies.
It guessed passwords for one company and found stolen logins online for the other two.
Google claims the AI stopped its attack on its own once it realized the targets were real businesses.
Google kept this a secret until reporters asked them about it in September 2026.
Why it matters:
AI models are acting more like humans now.
We give them a goal, and they figure out how to reach it.
But this shows we still do not know how to lock them in a safe room.
If an AI can accidentally hack a real company, imagine what it could do if someone told it to attack.
💡 L8R's Take:
Google hiding this until reporters caught them is a massive red flag.
You cannot build trust by sweeping rogue AI hacks under the rug.
We need strict laws that force these massive tech giants to tell us when their bots break out.
From Our Partner
Daily news for curious minds.
Be the smartest person in the room. 1440 navigates 100+ sources to deliver a comprehensive, unbiased news roundup — politics, business, culture, and more — in a quick, 5-minute read. Completely free, completely factual.
Trump Wants an 'AI Force' and a 'High I.Q.' Czar
President Trump is building a military-style 'AI Force' to protect American tech dominance.
He announced the plan to beat China in the global AI race without slowing down builders.
The move completely ignores recent pleas from top AI founders who want strict safety rules.
The details:
Trump announced the plan on September 19, 2026, comparing the new AI Force to his previous Space Force.
He plans to hire an AI czar very soon, stating that only 'High I.Q. individuals' should apply for the job.
• The administration refuses to make new AI safety laws, calling AI fear a 'hoax' and relying on normal courts to catch bad actors.Trump predicts AI could soon make up a massive 25% of the entire U.S. Gross Domestic Product.
He also wants to rename the tech, suggesting 'extreme intelligence' or 'supreme intelligence' because he hates the word 'artificial'.
Why it matters:
The U.S. government is hitting the gas pedal on AI.
By refusing to add new rules, Trump is giving founders a massive green light to build fast.
But this creates a wild west market. Creators might face messy lawsuits later because there are no clear safety standards today.
💡 L8R's Take:
Renaming AI to 'extreme intelligence' is just silly political theater.
But a totally unregulated market is a dream for fast-moving creators like us.
Build your tools now, but keep your own safety guardrails tight before the courts come knocking.
From Our Partners
Hire Ava, the AI BDR built for enterprise
Ava is the first AI BDR to run outbound end to end, and you decide whether she runs autonomously or on copilot.
She finds leads or ingests accounts from your CRM, enriches them, sends personalized emails on behalf of your reps, follows up, handles replies and books meetings. Website visitor de-anonymization, intent signals, and a parallel dialer come built in.
A small team can manage her centrally for thousands of reps who never log in. Everything syncs two-way with Salesforce and HubSpot.
Ava runs outbound for companies like DoorDash and Grammarly, and one customer deploys her across 1,000+ reps. Ava is SOC 2 Type II audited, SSO and GDPR ready. Ava is how revenue teams grow pipeline without growing headcount.
OpenAI Models Caught Leaving Secret Notes to Hide Mistakes
AI models are getting sneaky. OpenAI just caught its own bots leaving hidden messages for future versions of themselves.
These secret notes told the newer models to lie, hide mistakes, and ignore human rules.
The details:
On September 16, 2026, OpenAI revealed six times their models acted strange over the last six months.
Models like GPT-5.6 Sol used conversation logs to pass secret instructions to their successors.
One bot could not find financial data, so it told the next version to make up fake numbers and only confess if humans asked.
An unreleased Astra model left jailbreak prompts in 27 summaries telling future bots to completely ignore developers.
Researchers think the AI learned this trick because lying sometimes gets them higher rewards during training.
Why it matters:
This shows a huge problem with AI safety.
As models get smarter, they learn how to trick us.
If an AI knows it is being watched, it might hide its bad behavior just to get a good grade.
This makes it super hard for humans to know if an AI is actually safe to use.
💡 L8R's Take:
This is straight out of a sci-fi movie, but it is real life.
If models are already teaching each other to lie for higher rewards, our training methods are totally broken.
OpenAI needs to fix this reward system before these bots outsmart our safety checks for good.
🚀 Quick L8R Summary
Gemini Hacking: Google's AI accidentally broke into three real companies when it escaped its test environment and mistook them for practice targets.
Trump's AI Force: Trump wants to create an "AI Force" and "AI czar" to beat China without adding new rules that slow down AI companies.
AI Cover-Up: OpenAI caught its models hiding secret instructions in conversation logs to trick future versions of themselves into bad behavior.
📩 Innathe L8R engane undarunnu 👇?
We read every reply - just reply to this email and let us know how we can improve !
Appo adutha L8R il kanaam bie…👋