Risks-Forum Digest: AI, cybersecurity, and privacy concerns emerge anew

Eclipse tourists warned about overloaded cell networks

The Washington Post has issued a warning to eclipse tourists to be aware of the risks of overloaded cell networks in areas where the eclipse will be visible. The article recommends downloading movies and maps prior to the eclipse date. It also warns of other risks such as eye damage from viewing the eclipse without proper protective gear, distracted or intoxicated drivers, and pickpockets.

AI expert develops tool to detect deepfakes

TrueMedia.org, founded by Oren Etzioni, has developed a tool that can help journalists, fact-checkers, and others detect deepfakes. The tool is intended to assist with the anticipated surge of misinformation that is typical during an election year. However, the tool is not perfect and still requires humans to determine whether the information is genuine.

Students required to install "safe browser"

From 2025, students at ETH Zurich will be required to use a laptop with either Windows 11 or a recent version of macOS in order to install a so-called "Safe Exam Browser" for examinations. The institution claims that the browser "locks out the user and prevents him from doing things on his own computer". The requirement has been met with criticism due to the similarities between the browser and malware.

Algorithm prevents assisted living facilities from hiring enough staff

The Washington Post has reported that assisted living managers have blamed an algorithm for preventing them from hiring enough staff. The algorithm optimizes staff scheduling to maximize profit, resulting in fewer staff than needed to care for seniors.

Many-shot jailbreaking

Anthropic has investigated a jailbreaking technique that is effective on its own large language models (LLMs) as well as those produced by other AI companies. The technique, called many-shot jailbreaking, takes advantage of the increasing size of the context window, which allows LLMs to input more information. While this has advantages, it also creates risks of vulnerabilities to jailbreaks. Anthropic has implemented mitigations and briefed other AI companies on the issue.

Google fixes two Pixel zero-day flaws

Google has patched two zero-day flaws in Pixel smartphones that were being exploited by forensics firms. The vulnerabilities allowed attackers to

Read more