Open-source AlphaCodium brings back the adversarial concept to produce high integrity code and provides a path for Policy-as-code AI Security Systems

An uptick in LLM generated bounty submissions increases asymmetric costs to developers and is a systemic risk for the platforms

Timeline published, plus hotseat.ai brings Act to life

ChatGPT gains new guardian_tool - a policy enforcement tool

Anthropic: we don't know how to stop a model from doing the bad thing

Skeptical of current LLM benchmarks? There is another way...

Use generative AI to re-create missing library and components

Jetmo from UC Berkeley generates task specific LLMs

Security needs to weigh risk/reward before jumping to No

WAF meets MAP

Capture this in your threat model

The Big Guns get safer together

My favourite command line llm tool grows wings

Clickworkers are part of the AI supply chain. How to vet?

Weekly benchmarks of popular LLMs using real-world prompts

Researchers apply ML and AI to Side Channel Attacks

Always read the Eval label

Recordings now on YouTube

Novel and long tail failure modes of text-to-image models

Does your org manage its cloud storage tokens?

Truly private AI. Can it pay for itself?

3D browser render of LLM Visualization

A good watch on LLMs as Evaluators

Dream machines: it's a feature, not a bug

A proof-of-concept JavaScript tool to prevent IP address leakage in ChatGPT interactions.

AI Security: The Limits of Bug Bounty Programs and the Need for Non-ML Red Teaming

Enhancing LLM Output Predictability and Safety with Structured Validation

Emergent Challenges: Prompt Injections and Ensuring AI Security in an Unpredictable Landscape

Establishing Conversational Roles and Addressing Syntax-Level Prompt Injections

Have you agreed on a safe word with your loved ones yet?

Funny Jailbreak of the Week

LLMs as Colleagues? 8 Observations and Future Workplace Implications

Hallucinations as Bugs: AI's Double-edged Sword in Disruptive Technology and Society.

Building Trust in AI: Seeking Mechanistic Interpretability for AI Explainability and Safety

Evading AI Guardrails: Crafting Malware with ChatGPT's Assistance

AI-Powered Security: 7 Google Products Enhancing Protection

Tackling AI Safety & Alignment Challenges Amid Rapid Progress and Potential Disruptions.

AI-driven Socket identifies and analyzes 227 vulnerable or malicious packages in npm and PyPI repositories.

Scaling Supervision for Improved Transparency and Accountability in Reinforcement Learning from Human Feedback Systems.

A closed-loop learning system for enterprise Security Operations Centers

What is your organisation doing to control the potential downside of services like ChatGPT, whilst capturing the upside?

Insurance Companies Eye AI Risks: The Need for Employee AI Policies and Guardrails in Cybersecurity Management.

BadDiffusion Attack: Exposing Vulnerabilities in Image Generation AI Models and Exploring Risk Mitigation Techniques.

Debunking AI Model Theft Myths: Understanding Confidential Computing & Security Engineering in Modern GPUs.

GPT's Potential in Smart Contract Auditing: Current Limitations and Future Optimism as AI Capabilities Rapidly Improve.

Can an AI hide if its goals or objectives are not correctly aligned with those of its human designers or users (misalignment)?

AI-Assisted Cyberattacks: LLMs as Double-Edged Swords in Network Intrusions - Risks, Opportunities, and Detection Strategies.

OpenAI published a 60-page System Card, a document that describes their due diligence and risk management efforts

Can an AI be kept in its box?

AI Governance Dilemma: Leaked Llama Model Outperforms GPT-3! Explore the debate on trust, policy, and control as cutting-edge AI slips into public domain.

Expanding AI Threat Landscape: Untrusted Data Injection Attacks on Application-Integrated LLMs.

This MITRE ATLAS case study helps bring to life the framework

Facial recognition has lodged itself in people’s minds as the defacto technology for visual surveillance, and we should all find that quite disturbing!

“Co-appearance” sounds like a movie credit, but, in this case, you might not have signed up for the role.

If your organisation undertakes adversarial simulations, learn about ATLAS

Companies with proprietary source code can use public AI to generate regular and adversarial unit tests without disclosing their complete source code to said AI.

Can error messages from software teach an AI a new skill?

6 million people viewed post, but only one report an error by the AI

Testing strength of putting context in “System” vs “Messages” for ChatGPT

Tracking if your prompt or fine-tuned model is improving can be hard, but another LLM can judge the output of your model.

Enterprise companies are reportedly restricting their employees from using ChatGPT due to security and privacy concerns.

What is the likelihood of OpenAI facing data privacy enforcement under GDPR, according to privacy professionals?

AI safety research is woefully underfunded. Are we ready to manage the next existential risk since nuclear?

Four tactics and example prompts for hacking

Discover an unexpected failure mode of a superhuman AI system

Use a free voice creation service to impersonate a bank customer

Discover early stumbles of AI-enabled Bing and what it means for the future of AI.

What is the 'GPT-judge' automated metric introduced by Oxford and OpenAI researchers to evaluate model performance?

Adversaries can create 'BadNets' to misbehave on specific inputs, highlighting need for better neural network inspection techniques

What did a study by Oxford and OpenAI researchers reveal about the truthfulness of language models compared to human performance?

NIST warns: Integrated risk management essential for interconnectivity of AI, privacy, and cybersecurity risks.

AI can form and run a US LLC without humans, but with legal liability, security risks, and potential bias, should we grant it limited legal liability?

Entrepreneur used Deep Fake to send 10K thank you videos. Is this the first ethical use case for Deep Fake technology?

DAN (Do Anything Now) is the latest ChatGPT jailbreak, punishing the model for not answering questions

Researchers discover letter sequences that OpenAI's completion engine couldn't repeat, hallucinate, or complete correctly, leading to evasive responses.

Sophisticated techniques disrupt and steal Machine Learning models, but software and network vulnerabilities remain the biggest threat

NATO tested cyber defenders to maintain systems and power grids during a simulated cyberattack, with critical systems at risk

How did eight US marines evade detection by AI security cameras?

What are the focus areas of the US-EU partnership for AI research?

Nvidia's CSO describes how AI changes the dynamic between defenders and attackers

Take a security alert and use an AI to generate a remediation plan

Companies embracing AI must establish policies to label content amid complexities of determining AI-generated, mixed or biased content, and political considerations.

Malicious insider code is a significant security challenge. Can AI help?

A white paper suggests curriculum design emphasizing collaboration between humans and AI and raises questions about ChatGPT's potential to cheat internal training.

AI's Potential to Automate Web Browsing

ImaginAIry's image manipulation tool has use cases, but potential nefarious uses and detection concerns are worth noting.

Generate exploit code using an AI

An AI generated poem about Ransomeware

Learn how to create dynamic phishing campaigns in multiple languages with an AI

What traps should you avoid to get ahead using AI?

How a hacker reverse engineered an AI feature in Notion to reveal the underlying prompts