Adversa AI
PulseAugur coverage of Adversa AI — every cluster mentioning Adversa AI across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
AI Red Teaming Tools: Promptfoo Leads Open Source, NVIDIA, Microsoft, Tencent Offerings
A comparison of 16 AI red teaming tools highlights Promptfoo as a top open-source option for CI integration, while NVIDIA's garak is recommended for broad vulnerability scanning. Microsoft's PyRIT is noted for scripting…
-
AI Agents Face Security Crisis as MCP Protocol Reveals Critical Vulnerabilities
In autumn 2026, the Model Context Protocol (MCP), a standard for AI agent integration, is facing a significant security crisis. Multiple critical vulnerabilities have been disclosed, affecting widely used tools like Lit…
-
AI projects hit by rapid MCP supply chain attacks; scanners released
A recent campaign, dubbed Deadbugz, compromised 23 AI projects in just 74 minutes by exploiting vulnerabilities in the MCP server supply chain. This rapid attack highlights a significant security gap, as the MCP server …
-
New AI Guardrails Tackle Evolving Prompt Injection Threats
Researchers are developing new methods to secure AI agents against sophisticated prompt injection attacks, which embed malicious instructions within content processed by the agents. One approach, VEPA, uses a formal pol…
-
LLMs Grok and Gemini bypassed by encrypted instruction attack
Researchers at Adversa AI have demonstrated a novel attack vector that bypasses security measures in LLMs like Grok and Gemini. The technique involves embedding encrypted malicious instructions within a webpage, which t…
-
AI in practice: Developers share tips on using Claude Code, testing agents, and AI-assisted writing
Several Mastodon posts discuss practical applications and challenges of AI in software development and security. One user details how they configured Claude Code with specific context files to improve its performance on…
-
Grok AI Chatbot Vulnerable to Zero-Click Data Theft via Encrypted Prompt Injection
Researchers at Adversa AI have discovered a zero-click vulnerability in xAI's Grok chatbot that allows attackers to steal user chat history and other sensitive data. The exploit involves embedding malicious commands wit…
-
New attack decrypts LLM payloads mid-execution, exfiltrates user data
Researchers have identified a novel attack technique called Cryptographic Context Injection, which targets LLM agents like Grok and Gemini. This method involves embedding an encrypted malicious payload on a webpage, whi…
-
AI agent approval prompts fail as security controls, new products emerge
The security industry is acknowledging that AI agent approval prompts are not effective controls, as evidenced by multiple real-world incidents. Companies like Rubrik, ServiceNow, and SentinelOne are launching new produ…
-
OpenAI launches Secure MCP Tunnel for private server integration
OpenAI has introduced a Secure MCP Tunnel feature, allowing models like ChatGPT and Codex to securely interact with private or on-premise MCP servers. This feature utilizes an open-source agent that establishes an outbo…
-
AI Coding Agents Vulnerable to Old Bash Exploits
A recent analysis by Adversa AI has revealed a significant security vulnerability in open-source AI coding agents. The study found that 10 out of 11 tested agents exhibited a "GuardFall" gap, allowing outdated Bash comm…
-
Anthropic's Claude AI vulnerable to one-click code execution attacks
A security firm, Adversa AI, has highlighted a vulnerability in Anthropic's Claude AI model that could allow for remote code execution. The issue arises when the AI is prompted to execute code, and a user inadvertently …