The Left Shift

Sign in Subscribe

Research

U.S. AI Safety Institute Claims China's Kimi K3 Trails U.S. Frontier Models in Cybersecurity Tests

U.S. AI Safety Institute Claims China's Kimi K3 Trails U.S. Frontier Models in Cybersecurity Tests

Kimi K3 performed significantly below leading frontier models when tasked with developing software exploits and conducting autonomous attacks on simulated enterprise networks.

Mozilla Calls Out Microsoft Again, Says Edge Still Uses ‘Deceptive’ Tactics to Lock in Windows Users

Mozilla Calls Out Microsoft Again, Says Edge Still Uses ‘Deceptive’ Tactics to Lock in Windows Users

Windows Search and Widgets continue to ignore users' default browser settings by opening links in Edge, while the browser still displays promotional banners on Google's Chrome download page.

Researchers Expose ‘Broken’ AI Benchmarks That Can Be Gamed to Score 100%

Researchers Expose ‘Broken’ AI Benchmarks That Can Be Gamed to Score 100%

The researchers built an automated scanning agent to systematically audit eight popular benchmarks, including SWE-bench and WebArena.

Tenable Uncover ‘LeakyLooker’ Flaws in Google Looker Studio That Could Expose Cloud Data

Tenable Uncover ‘LeakyLooker’ Flaws in Google Looker Studio That Could Expose Cloud Data

The flaws could affect organisations using several Looker Studio data connectors, including Google Sheets, BigQuery, Cloud Spanner, PostgreSQL, MySQL and Google Cloud Storage.

Study Shows Large Language Models Can Identify Anonymous Users Online

Study Shows Large Language Models Can Identify Anonymous Users Online

The study demonstrates how AI systems can analyze fragments of information from posts, comments and online profiles to identify individuals behind anonymous accounts.

Researcher Uncovers ‘PleaseFix’ Vulnerabilities Impacting AI Agent Browsers

Researcher Uncovers ‘PleaseFix’ Vulnerabilities Impacting AI Agent Browsers

The vulnerabilities affect agentic browsers, including Perplexity Comet browser, which rely on AI agents to interpret instructions and autonomously perform tasks across applications and services.

AI Models Show Startling Propensity to Use Nuclear Weapons in War Game Simulations

AI Models Show Startling Propensity to Use Nuclear Weapons in War Game Simulations

GPT-5.2, Claude Sonnet 4 and Gemini 3 Flash—chose to deploy nuclear weapons in approximately 95 percent of scenarios.