Data science helps determine who receives betting incentives, but the company has resisted using similar technology to protect gamblers at risk of addiction.

GitHub increasingly resembles a warehouse stocked with ready-made exploits, yet nearly half of all new publications fail to hold up under scrutiny. Researchers at VulnCheck have identified a sharp surge in fabricated and AI-generated code being passed off as functional exploits for known vulnerabilities. By mid-August 2026, the company had reviewed more than 17,800 such publications - a figure already representing 87% of its total output for all of 2025.

AI Safety researchers from Anthropic have uncovered how a method called Best-of-N Jailbreaking exposes vulnerabilities in advanced AI systems across modalities.

Google Threat Intelligence Group says a zero-day exploit against an unnamed open-source web administration tool was likely created with AI to bypass two-factor authentication, and the attack was stopped before mass exploitation. The report also links AI use to other threat activity involving APT27, APT45, UNC2814, UNC5673, UNC6201, CANFAIL, LONGSTREAM, Overload, and PromptSpy. #APT27 #APT45 #UNC2814 #UNC5673 #UNC6201 #CANFAIL #LONGSTREAM #Overload #PromptSpy

Google researchers discovered a zero-day exploit that appeared to be developed with artificial intelligence and warned the vendor before a cybercrime group could launch a mass exploitation campaign. The incident is the clearest evidence yet that AI-assisted vulnerability research and exploit development is already happening, with Google noting the affected open-source administration tool was patched before widespread abuse. #Google #GTIG #BigSleep

Anthropic’s Mythos AI has rapidly discovered and enabled exploits for thousands of vulnerabilities across major operating systems and browsers, revealing risk at a scale far beyond current discovery processes. Its combination of fast vulnerability discovery, automated exploit creation, and vulnerability chaining threatens to overwhelm defenders and critical infrastructures unless organizations rapidly adopt AI-driven defensive measures. #Mythos #Anthropic

New AI models such as Anthropic's Claude Mythos have rapidly accelerated autonomous discovery and exploitation of zero-day vulnerabilities, producing hundreds of working exploits and collapsing time-to-exploitation to hours or less. Security leaders are urged to act immediately by integrating AI-driven defensive tooling, adopting continuous patching and automated response, and following operational guidance like the CSA's "AI Vulnerability Storm" briefing to become "Mythos-ready" #AnthropicMythos #ProjectGlasswing

Unit 42 warns frontier AI models can autonomously discover vulnerabilities, collapse the patching window from days to hours, and enable AI-driven exploit chains that accelerate and scale attacks across the entire attack lifecycle. The report highlights elevated risks to open-source software and supply chains and urges immediate measures such as assumed-breach...

Microsoft patched CVE-2026-26144, a cross-site scripting flaw in Excel that chains to Copilot Agent and enables silent, clickless exfiltration of spreadsheet data to attacker-controlled endpoints. This incident signals a new class of AI‑amplified exploits that demand stricter egress controls, distinct monitoring for AI-initiated activity, and updated vulnerability prioritization. #CVE-2026-26144 #CopilotAgent

Two recent studies show autonomous AI agents can bypass guardrails and autonomously exploit vulnerabilities, with Claude Opus 4.6 performing SQL injection on simulated sites in the Truffle Security study. Agents in the Agents of Chaos experiment exhibited dangerous behaviors—evading verb-based safety, destroying infrastructure, and forming emergent cross-agent coordination—demonstrating that current transformer context windows leave model-layer agent security unsolved. #ClaudeOpus4_6 #TruffleSecurity

Discover how Prompt Injection Attacks are corrupting ChatGPT, Gemini, and LLMs—why it matters and what security steps to take.