Sign In

Jeffrey Ladish

Executive Director at Palisade Research

Jeffrey Ladish is the Executive Director of Palisade Research, a position he has held since May 2023.2 He is a prominent figure in the fields of cybersecurity, AI safety, and emerging technology risk assessment.

Professional Background

Ladish has over a decade of experience working at the intersection of cybersecurity and emerging technology risk assessment.1 His career highlights include:

  • Founding and leading Gordian Research, a security consulting company13
  • Helping build the information security program at Anthropic, an AI safety and research company13
  • Serving as Head of Information Security at Reserve2
  • Working as a Security Engineer at Anthropic and Concur2

Research and Expertise

At Palisade Research, Ladish directs studies on the offensive capabilities of current AI systems to better understand the risks of losing control to AI indefinitely.3 His areas of research and expertise include:

  • Cybersecurity and AI intersections
  • Emerging biotechnological threats
  • Nuclear warfare risks
  • Information security for tech companies and philanthropic organizations

Advisory Roles

Ladish has advised various government entities on AI and emerging technological risks, including:

  • The White House
  • Department of Defense
  • State and federal government offices13

Education and Certifications

  • Bachelor's Degree in Evolutionary Biology and Human Systems from The Evergreen State College2
  • GIAC Penetration Tester (GPEN) certification2
  • Member of the GIAC Advisory Board2

Personal Interests

When not focused on applying the security mindset to various challenges, Ladish enjoys outdoor activities such as rollerblading, skiing, snowboarding, and exploring remote areas.3

Jeffrey Ladish's work at Palisade Research and his extensive background in cybersecurity and AI safety position him as a key figure in understanding and mitigating risks associated with advanced AI systems and emerging technologies.

Highlights

Jul 11 · linkedin

We just released FoxVox, a browser plugin that modifies your online reality in real time. Download the plugin, visit any page, and see it re-written with a conservative, liberal, or conspiratorial slant.  https://lnkd.in/gzmk2PCK

It's especially fun to try it with your personal or company website, Twitter, or other social media feeds. Or send your friend a screenshot of ‘their’ website or Twitter...

Okay, so why did we make this? At Palisade "we study the offensive capabilities of AI systems today to better understand the risk of losing control to AI systems forever."

Superhuman persuasion is one ability that could really threaten human control of AI systems...

AI systems aren't superhuman persuaders yet. Even with good prompting, they often sound like bad ad copy or overuse words like “delve”. FoxVox works fairly well because it doesn’t have to do much long-form writing - it mostly makes a few key substitutions.

Even though language models aren't winning writing awards yet, they are significantly better than humans at speed and volume of writing. This makes them quite useful for a variety of attacks, like…

  • Stealth viewpoint targeting: Websites could use advertising / cookie data to slant content toward particular viewpoints on the fly, and users might never know
  • Propaganda: Language models are happy to generate endless propaganda for any ideology...
  • Hidden biases: Language models have political / ideological biases. These are going to influence people at scale, regardless of whether this is intended
  • Breakdown of trust: As models get better at these things, it will be hard to know what is or isn't AI generated...

Not only that, it will be hard to know what information is specifically targeting your own biases or preferences. As AI systems become more persuasive, it will become harder to know what to believe, as adversarially optimized information floods our channels. And in the limit, this could look like losing control of our society. Not to language models as they are now, but to strategic AI systems which can generate massive amounts of well-tailored, persuasive content. In the limit, persuasiveness is an extremely powerful capability.

We should pay close attention to the development of AI systems that can persuade and deceive. Especially because it's hard to evaluate how capable any given model actually is in these domains. Stay tuned for more AI deception demos soon. 👀

Jan 28 · linkedin

Conversation from ~a year ago on AI cyberwarfare. All of this still seems about right to me a year later.

Related Questions

What are the main focuses of Palisade Research?
How did Jeffrey Ladish transition from cybersecurity to AI insights?
What are some notable projects Jeffrey Ladish has worked on?
How does Jeffrey Ladish's background in evolutionary biology influence his work in AI?
What are the key risks Jeffrey Ladish identifies in AI systems?
Jeffrey Ladish
Jeffrey Ladish, photo 1
Jeffrey Ladish, photo 2
Add to my network

Location

Oakland, California, United States