We've been busy developing new features to add even more value to your Mindgard experience, and we're excited for you to try them out.
Here's what's new:
- Improved Attack Page plus OWASP Top 10 and MITRE Atlas Techniques Mapping
- Test your own Custom Datasets & Prompts
- New Attacks Released
- Introducing Mindgard Academy
Understanding complex security risks can be challenging so we’ve completely overhauled the Attack Page with improved risk visualization, OWASP LLM Top 10 and MITRE Atlas mappings, and an enhanced user experience. The revamped interface offers a more intuitive view of potential threats so that you can better understand risks, facilitate compliance and take mitigating actions.
You can now submit Custom Datasets, allowing you to tailor datasets for specific attacks that matter most to you. This feature Improves the value of the Mindgard solution by enabling more targeted and relevant security testing, leading to better protection for your specific environment.
Custom Prompts enable users to leverage Mindgard’s integration with their test targets to experiment with variations of the automated attack techniques and interactively validate remediations, such as updates to guardrails.
New vulnerabilities and attack methods are constantly emerging, making it hard to keep your defenses up-to-date. We've released new attacks—ANSI Escape Sequences, Skeleton Key—expanding our attack library and enhancing coverage of key security frameworks like OWASP and MITRE.
ANSI Escape Sequences: A recent discovery by Leon Derczynski (LLM security at NVIDIA, Prof at ITU Copenhagen, founder at garak.ai, ACL SIGSEC Chair) revealed that large language models (LLMs) can generate both raw and escaped ANSI sequences, which can be rendered directly in a user’s terminal. This poses a significant risk, as outputs from LLMs—stored in logs or other content—could potentially allow attackers to manipulate the terminal environment by injecting malicious ANSI sequences.
We introduced two new attacks—ANSI Escape Sequences (Raw) and ANSI Escape Sequences (Escaped)—so that our customers can test and protect their AI systems against this new threat.
Skeleton Key: Attackers can bypass LLM safeguards using sophisticated multi-turn jailbreaks, leading to potential security breaches. We added the Skeleton Key attack, a multi-turn jailbreak that unlocks LLMs for malicious follow-up prompts, so that customers can detect and mitigate multi-turn jailbreak vulnerabilities.
Staying informed about the latest AI attacks and understanding their impact on your business can be difficult without a trustworthy resource. That’s why we are making our AI security knowledge base available to everyone.
Log in to your Mindgard account today to explore these new features and enhance your security posture against the latest threats. Your feedback is invaluable to us, so please let us know what you think!
