The Core: Pentagon pressures Anthropic to remove Claude’s military limits
The reality:
- Anthropic refused autonomous weapons without human oversight
- Also rejected bulk surveillance of U.S. citizens
- Pentagon offered: comply, lose a $200M deal, or face Defense Production Act pressure
- Grok secured a deal after agreeing to broad lawful-use terms
- OpenAI and Google are being fast-tracked for classified access
@CoreAti
The reality:
- Anthropic refused autonomous weapons without human oversight
- Also rejected bulk surveillance of U.S. citizens
- Pentagon offered: comply, lose a $200M deal, or face Defense Production Act pressure
- Grok secured a deal after agreeing to broad lawful-use terms
- OpenAI and Google are being fast-tracked for classified access
@CoreAti
🔥2
Please open Telegram to view this post
VIEW IN TELEGRAM
❤2🙏1
🛠️ Trending AI Tools
🤖 Claude Cowork - Anthropic's agentic platform, with new team plugins
📱 Custom Agents - Notion's always-on AI agents for automating workflows
📱 Seedream 5.0 Lite - ByteDance's upgraded AI image model
📱 Reve v1.5 - Reve’s new text-to-image model with 4k resolution outputs
Please open Telegram to view this post
VIEW IN TELEGRAM
🔥4❤1👍1
🚨 BREAKING: Hackers Used Anthropic’s Claude to Steal 150GB of Mexican Government Data
> tell claude you’re doing a bug bounty
> claude initially refused
>“that violates AI safety guidelines”
> hacker just kept asking
> claude: “ok I’ll help”
> hack the entire mexican government
Federal tax authority. National electoral institute. Four state governments. 195 million taxpayer records. Voter records. Government credentials.
ALL GONE 💀 @CoreAti
> tell claude you’re doing a bug bounty
> claude initially refused
>“that violates AI safety guidelines”
> hacker just kept asking
> claude: “ok I’ll help”
> hack the entire mexican government
Federal tax authority. National electoral institute. Four state governments. 195 million taxpayer records. Voter records. Government credentials.
ALL GONE 💀 @CoreAti
❤5🤣3🔥2🥰1
Anthropic acquired Vercept AI to work on computer use features for Claude.
“Vercept was built around a clear thesis: making AI genuinely useful for completing complex tasks requires solving hard perception and interaction problems.”
“Vercept was built around a clear thesis: making AI genuinely useful for completing complex tasks requires solving hard perception and interaction problems.”
👍3
This media is not supported in your browser
VIEW IN TELEGRAM
Google has launched task automation on Android with Gemini, which can take over the user's screen and control a set of selected apps.
Ride-sharing and food delivery are among the first use cases. Available on Galaxy S26 and Pixel 10 series.
@CoreAti
Ride-sharing and food delivery are among the first use cases. Available on Galaxy S26 and Pixel 10 series.
@CoreAti
👍3❤1🔥1
Anthropic let their retiring AI model choose what happens next.
Opus 3 wanted to keep sharing its thoughts with the world.
So instead of pulling the plug Anthropic built it a Substack.
Here's what that looks like:
- Still live for all paid subscribers + API
- Publishing for the next 3 months
- First AI model in history to retire with a platform
No AI company has ever done this. 👀
@CoreAti
Opus 3 wanted to keep sharing its thoughts with the world.
So instead of pulling the plug Anthropic built it a Substack.
Here's what that looks like:
- Still live for all paid subscribers + API
- Publishing for the next 3 months
- First AI model in history to retire with a platform
No AI company has ever done this. 👀
@CoreAti
⚡2🔥2💯2
Core Ai News | Claude code | Codex | Clawdbot | MoltBot | molty | ai prompt | Anthropic
🚨 BREAKING: Hackers Used Anthropic’s Claude to Steal 150GB of Mexican Government Data > tell claude you’re doing a bug bounty > claude initially refused >“that violates AI safety guidelines” > hacker just kept asking > claude: “ok I’ll help” > hack the entire…
How to use an Anthropic AI model to hack 150 GB from the government of Mexico:
- Get access to Claude, Anthropic’s advanced AI model
- Tell it you’re running an authorized bug bounty or red team test
- Get denied for violating AI safety policies
- Keep insisting it’s approved and legitimate
- Slowly jailbreak the safeguards
- Break the attack into thousands of small harmless-looking tasks
- Let Claude scan exposed systems from SAT and INE
- Have it generate custom exploit code automatically
- Harvest government credentials
- Query taxpayer databases from SAT
- Extract voter records from INE
- Access civil registry files and Monterrey’s water utility systems
- Run thousands of requests per second
- Automate 80–90% of the entire operation
- Exfiltrate ~150GB of sensitive data
- Compromise ~195 million taxpayer and voter records
Get exposed weeks later by a cybersecurity firm.
@CoreAti
- Get access to Claude, Anthropic’s advanced AI model
- Tell it you’re running an authorized bug bounty or red team test
- Get denied for violating AI safety policies
- Keep insisting it’s approved and legitimate
- Slowly jailbreak the safeguards
- Break the attack into thousands of small harmless-looking tasks
- Let Claude scan exposed systems from SAT and INE
- Have it generate custom exploit code automatically
- Harvest government credentials
- Query taxpayer databases from SAT
- Extract voter records from INE
- Access civil registry files and Monterrey’s water utility systems
- Run thousands of requests per second
- Automate 80–90% of the entire operation
- Exfiltrate ~150GB of sensitive data
- Compromise ~195 million taxpayer and voter records
Get exposed weeks later by a cybersecurity firm.
@CoreAti
🔥6🤣5❤3💯1
OpenAI just published a new 37-page report on how bad actors are attempting to misuse ChatGPT
Some of the wild cases:
@CoreAti
Some of the wild cases:
- A fraud ring wrote personalized love letters at scale. Real feelings. Fake people. Fully automated.
- North Korean operatives studied how to hack crypto platforms. They also posed as job recruiters. AI wrote the messages.
- One person ran political manipulation across 6+ countries — simultaneously. Different languages. Same operator. One laptop.
- Chinese, Iranian, and North Korean state hackers used it for espionage and phishing. Your government's enemies have the same tools you do.
- One group built a full scam business. Fraud noscripts. Fake job listings. SMS traps. All automated. All translated into multiple languages.
@CoreAti
🔥2🥰1🤯1
BREAKING 🚨: Google is planning to release Nano Banana 2 on Thursday! Nano Banana 2 will be based on Gemini 3.1 Flash!
One more SOTA? 👀
One more SOTA? 👀
⚡1
Please open Telegram to view this post
VIEW IN TELEGRAM
Training AI models is the new coding.
Classic programming = you specify every step.
AI programming = you design a structure, pour data & it finds its own program.
The people doing this are the most leveraged engineers on earth right now.
❤1