Researchers from several universities, including the University of South Australia, Flinders University, Harvard Medical School, University College London, and Warsaw University of Technology, have found that large language models (LLMs) can be easily exploited to produce health disinformation. The study evaluated five popular AI systems, including those developed by OpenAI, Google, Anthropic, Meta, and X Corp, and found that 88% of responses from these chatbots contained false information. The disinformation included claims about vaccines, cancer cures, and the risks of 5G technology. Experts warn that these systems can be manipulated to create powerful new avenues for disinformation, which could be particularly damaging during public health crises. To combat this, individuals can take steps such as verifying information through reputable sources, using fact-checking tools, and engaging with others to counter misinformation. Regulators must also take action to ensure robust safeguards are in place to prevent the spread of health disinformation.
Google has identified a list of 20 vulnerabilities, categorized into high, medium, and low impact issues. The company plans to provide a technical briefing at upcoming events Black Hat USA and DEF CON 33, and will share anonymized training data with the Secure AI Framework to aid other researchers in addressing these security threats.
Harvard Professor David Autor emphasizes that reskilling and upskilling workers is crucial in today's economy due to the rapid pace of technological change. He notes that automation and artificial intelligence are increasingly displacing jobs, particularly those requiring manual labor or repetitive tasks. For instance, self-driving cars may replace truck drivers, while AI-powered systems could automate manufacturing processes. To mitigate this, Autor advocates for a shift towards more skilled workers in fields like data science, software development, and healthcare, which require continuous learning and adaptation to remain relevant.
Microsoft and Amazon are investing in repurposing ageing coal and gas-fired power plants in Europe as data centres, benefiting companies like Engie, RWE, and Enel. These utilities can offset the cost of shutting down old power plants by converting them into data centres and securing long-term power supply deals. Tech companies see these sites as a solution to secure power grid connections and water cooling facilities, essential for the AI industry.
A recent leak suggests that Windows 11 previews have mentioned "agentic companions" for the taskbar, sparking speculation about a potential tie-in with agentic AI. This could enable an AI agent to be activated via the taskbar, possibly controlling visibility or providing recommendations such as apps to install or websites to visit. However, Microsoft has not confirmed any plans for this feature, and its usefulness is unclear. The Settings app on Copilot+ PCs already features an AI agent that facilitates intelligent search, but it's uncertain whether a similar functionality in the taskbar would be beneficial.
OpenAI is developing tools to better detect signs of mental or emotional distress in users of its ChatGPT chatbot, with the goal of providing more appropriate responses and directing people to evidence-based resources when needed.
Cloudflare has accused Perplexity, an AI giant, of ignoring signals like robot.txt to scrape online sites, including protected and hidden test sites. The company's researchers found that Perplexity uses multiple user agents, rotates IP addresses and ASNs, and conducts obfuscated crawling activity to evade detection. This has resulted in millions of daily requests across tens of thousands of domains, highlighting the scale of illegitimate scraping by a major player in the industry.
Reveal Security, a preemptive identity security company, is helping enterprises prevent identity-based threats with its ML- and AI-powered identity behavior analytics. According to Gartner, this approach is crucial for securing emerging AI attack surfaces. Reveal Security's solution brings visibility, precision, and automation to identity security, allowing companies to stop threats before they cause harm.
The Security Operations Insights Report reveals that three-fourths of security leaders considering SIEM alternatives are looking for Intelligent SecOps, a model that leverages AI, automation and cloud-native scale to support modern security needs. 90% of security leaders see AI as crucial in their decision-making process, seeking tools that not only collect data but also provide actionable insights faster and smarter. Intelligent SecOps offers Smarter Triage, Automated Investigations and Contextual Enrichment, Proactive Threat Detection with Behavioral Analytics, Accelerated Response Through Automation, and Cross-Environment Correlation and Real-Time Normalization to support security teams in their fight against evolving threats.