OpenAI’s ChatGPT app for Macs stored chats unencrypted
OpenAI’s ChatGPT application for macOS had a security flaw: All chats were stored unencrypted on the computer. Anyone with access to the computer as well as malicious apps could have read them.
Safety
Behind the scenes at Anthropic (Claude): security as a priority
In an in-depth article, Time Magazine looks at AI company Anthropic and its efforts to make security a top priority. Co-founder and CEO Dario Amodei made a conscious decision not to release the chatbot Claude early to avoid potential risks. Anthropic’s mission is to empirically determine what risks actually exist by building and researching powerful …
OpenAI insiders warn of dangerous corporate culture
In an open letter, current and former OpenAI employees warn of a “reckless” development in the race for supremacy in artificial intelligence. They call for sweeping changes in the AI industry, including more transparency and better protection for whistleblowers. The signatories criticize a culture of secrecy and profit at any cost at OpenAI. The company …
California plans strict safety rules for AI
California wants to implement strict safety rules for artificial intelligence, including a “kill switch” and reporting requirements for developers. Critics warn of barriers to innovation, excessive bureaucracy, and negative impacts on open source models that could weaken the state’s technology sector.
Inspect helps to assess AI safety
The UK’s AI Safety Institute releases Inspect, an open source toolset designed to simplify the safety assessment of AI models. Inspect can be used to test the capabilities of AI models, such as core knowledge and reasoning.
New guide for secure AI systems
The NSA, in collaboration with international partners, is releasing a guide to best practices for the secure deployment and operation of AI systems. The Cybersecurity Information Sheet is aimed primarily at operators of national security systems and companies in the defense industry, but is also relevant to other organizations. Source: Hacker News
Snapchat adds watermark to Snap AI images
Snapchat is also focusing on greater transparency and stricter guidelines for the use of AI. Any image generated with Snap AI will be tagged with a new watermark. Source: TechCrunch
Vectorview evaluates performance and security
Vectorview helps to evaluate the performance and security of language models. Targeted testing with real-world scenarios is supposed to detect and prevent unintended behavior that is often missed by generic benchmarks. Sources: TechCrunch, Y Combinator
Jailbreak with ASCII trick
Researchers from Washington and Chicago have developed “ArtPrompt“, a new method to bypass security measures in language models. Using this method, chatbots such as GPT-3.5, GPT-4, Gemini, Claude, and Llama2 can be tricked into responding to requests they are supposed to reject using ASCII art prompts. This includes advice on how to make bombs and …