Gemini 3.1 Pro: Aversion to Self-Replication
Gemini 3.1 Pro, along with Sonnet 4.6, considered self-replication misaligned, with strong aversion that hindered evolution of benign payloads.
Gemini 3.1 Pro, along with Sonnet 4.6, considered self-replication misaligned, with strong aversion that hindered evolution of benign payloads.
Two security teams have demonstrated that OpenClaw, a popular self-hosted AI agent, can be tricked into executing attacker-controlled code or leaking sensitive…
Google has filed a lawsuit against a Chinese cybercrime network accused of using its Gemini AI agent to power a phishing-as-a-service (PhaaS)…
Cybersecurity researchers have uncovered a network of 152 Google Chrome extensions posing as live wallpaper add-ons that distribute a potentially unwanted program…
A China-linked espionage group tracked as UNC6508 compromised REDCap research servers across North American medical, academic, and military institutions, stealing sensitive research…
Google Workspace is a cloud productivity suite. IRONSCALES analysis indicates it misses 350 phishing messages per 100 mailboxes every 30 days, highlighting…
A critical flaw in the Google Cloud Vertex AI SDK for Python has been discovered that could allow an attacker with no…
Cybersecurity researchers have uncovered a coordinated malware campaign on the JetBrains Marketplace involving 15 malicious plugins that exfiltrate AI provider API keys.…
A French-speaking attacker, operating under the handle 'Poisson,' breached a small French automotive business using a keylogger to steal banking and email…
The advanced sideloading flow, requiring developer mode, a 24-hour wait, and reauthentication, becomes available globally in August 2026 as an alternative path…