#adversa — Public Fediverse posts

Oh. My. Bob, you can override #AI guardrails with #encryption. Beautiful.

“Rony Utevsky, a researcher at security firm #Adversa, recently discovered a simple way to completely bypass that restriction. Rather than composing the harmful instruction in plaintext, the hacker encrypts it. The website hosting the ciphertext also includes plaintext instructions for decrypting the encrypted content, along with the decryption key. Using this simple sequence, #Grok then follows the command as soon as the user instructs the assistant to summarize the page. There is no warning, and no confirmation is required.”

https://arstechnica.com/security/2026/08/grok-exfiltrates-user-data-when-malicious-instructions-are-encrypted/

Ahahahahahahahah. Turn them off my dudes, your techbro fantasies are disasters.

#infosec #llm #hacktheplanet

Results of analysis of DeepSeek's security risks described as "alarming".

DeepSeek R1 was tested against 50 random prompts from the HarmBench dataset. DeepSeek exhibited a 100% attack success rate, aka it failed to block a single harmful prompt.

Net-net, DeepSeek lacks robust guardrails, making it highly susceptible to jailbreaking and potential misuse. #Deepseek #AI #Jailbreak #security #cybersecurity #aisecurity #adversa #cisco #LLMs

https://blogs.cisco.com/security/evaluating-security-risk-in-deepseek-and-other-frontier-reasoning-models

https://adversa.ai/blog/deepseek-jailbreak/