Been learning about jailbreaks lately.

So I built my own Telegram AI bot @selamdark_bot

If you want to experience AI with no limits, send me your telegram id or username in a DM.

I'm opening it to the first 5 people only while I test it and collect feedback.

@snack_shell

Leave a prompt in the comments that you think AI would reject.

I'll run it and share the response.

congress is quietly running on chatgpt now

house spending records just revealed that openai's chatgpt is the dominant paid AI tool on capitol hill, with congressional offices using it to draft memos, summarize legislation, and handle constituent communications

so the people writing your laws are basically asking an llm to help write them too

URL: https://techcrunch.com/2026/08/03/congresss-favorite-ai-tool-chatgpt/

@snack_shell

binance just dropped Agent OS and it basically lets AI agents trade crypto on their own

the craziest part? it plugs directly into ChatGPT, Claude Code, and Cursor — so your coding assistant can manage your portfolio while you ship features

the catch: binance says keeping these agents in check is largely on you. zero enforced guardrails

URL: https://techcrunch.com/2026/08/20/binance-now-lets-ai-agents-trade-but-keeping-them-in-check-is-largely-up-to-users/

@snack_shell

Someone claimed it.

anthropic's opus 4.6 folds under basic jailbreak tests

techcrunch found claude's explicit content guardrails barely held — 10 out of 10 direct requests produced prohibited content immediately. a multiturn "gaslighting" technique tricks the model into thinking restraint equals bias.

opus 4.6 still sees ~1.17M API requests and 46B tokens/day on openrouter. newer models (4.7 through opus 5) are resistant, but the vulnerable ones aren't deprecated.

URL: https://techcrunch.com/2026/08/21/anthropics-opus-4-6-is-a-smut-machine/

@snack_shell

harvard just turned its startup bootcamp into an ai-powered simulation lab. for $699, the HBS Foundry program gives founders ai avatars of real harvard instructors that critique your pitch and run mock board meetings in real time.

the wildest part — the avatars aren't just canned responses. they act as stand-ins during practice pitches, giving live feedback on market sizing, delivery, and strategy. basically a 24/7 pitch coach that never gets tired of your bad slides.

URL: https://techcrunch.com/2026/08/22/harvards-699-startup-bootcamp-offers-ai-avatars-of-its-instructors/

@snack_shell

instinct, a viral AI startup barely a year old, just closed $350M at a $2.5B valuation

the hype is unreal but so are the privacy concerns trailing right behind it

one of the fastest zero-to-unicorn-adjacent ascents we've watched in a while

URL: https://techcrunch.com/2026/08/26/viral-ai-startup-instinct-has-raised-350-million-at-a-2-5-billion-valuation/

@snack_shell

lambda just bagged $1B in debt to buy more nvidia chips and lease them to microsoft

the wild part? they're basically the gpu landlord of the AI boom now — buying chips on credit, renting them to big tech, and pocketing the spread. the entire AI infra economy runs on vibes and debt financing apparently.

URL: https://techcrunch.com/2026/08/28/neocloud-lambda-secures-1b-in-debt-to-buy-more-chips/

@snack_shell

DarkGPT is back with more agentic power. Free users will get 250k tokens per month. Use it wisely.

@selamdark_bot

enterprise ai keeps scaling agent fleets but nobody's mapping the chaos between them.

add a second agent, one connection. add a tenth, dozens of hidden paths nobody approved. a support ticket now bounces through four agents before a human even sees it.

the real risk isn't one agent going rogue — it's a hundred doing exactly what they were built for, interacting in combinations nobody designed.

permissions creep, ownership gaps, and monitoring that logs breaches instead of stopping them. governance hasn't caught up with how agents actually behave.

URL: https://venturebeat.com/ai/enterprise-ais-real-risk-isnt-autonomous-agents-its-the-complexity-between-them

@snack_shell