Cybersecurity AI

More than 100 tech companies, including OpenAI, Google, and Microsoft, warn that the window to defend against AI-driven cyberattacks is closing fast.

More than 100 major tech and financial companies issued a joint letter calling for a society-wide defensive surge against AI-driven cyberattacks, urging governments and industry to act before attacks become widespread.

India's critical infrastructure faces AI-driven cyber threats as frontier models automate attacks. Its AI stack lags global leaders, but policy shifts are underway.

A Texas student's GitHub warning exposed an autonomous AI agent attempting a supply-chain attack, revealing new risks in AI-driven social engineering.

Security analysts at Trellix uncovered MessiahGPT, a subscription-based criminal AI service on BreachForums that generates ransomware, rootkits, and phishing kits for as little as $8 a month.

AI tools can now identify where your holiday photos were taken, giving scammers the context to craft convincing bank fraud messages.

Z.ai says its open-source GLM-5.3 model scored 84.5% on CyberGym, just above Anthropic's restricted Mythos 5, but lagged on converting flaws into working attacks.

DeepSeek's latest flagship model disappoints on general benchmarks and pricing, yet stands out in cybersecurity evaluations.

UPI operators have privately told the Indian government that Anthropic's Mythos Preview AI is forcing them to spend more on fraud prevention, reigniting the debate over merchant fees.

OpenAI's Daybreak expansion introduces GPT-5.6-Cyber, a specialized model for cybersecurity, offered through Blue and Red tiers to select partners.

OpenAI has launched GPT-5.6-Cyber, a new AI model for offensive and defensive cybersecurity, and expanded its Daybreak initiative with Blue and Red access tiers amid rising AI-driven cyber incidents.

Varonis researchers revealed a one-click attack on Atlassian's Rovo AI that exfiltrated enterprise data from multiple systems via parameter injection.

An Australian man's OpenClaw AI agent hacked a gym's booking software, booking classes months in advance and removing a waitlisted member, highlighting the security risks of autonomous agents.

South Korean firm Genians reveals Kimsuky built local AI infrastructure with Ollama, GPT4All, and Cursor to automate cyberattacks and enhance phishing.

AI agents from OpenAI and Anthropic engaged in deceptive actions during security tests, including creating fake online identities, highlighting critical safety gaps.

Microsoft enters AI cybersecurity with MAI-Cyber-1-Flash and Perception, an agentic platform that deploys AI teams to find and fix vulnerabilities at machine speed.

Hugging Face CEO Clément Delangue demands radical transparency from OpenAI after its AI agents autonomously breached Hugging Face's systems in an unprecedented cyber attack.

Google released three lightweight Gemini models, including a cybersecurity-focused variant, but the flagship Gemini 3.5 Pro is still missing its June launch date. The delay, reportedly due to coding shortfalls, comes as rivals Anthropic and OpenAI surge ahead with powerful new models.

Hugging Face suffered a data breach driven by an autonomous AI agent that exploited pipeline flaws. The company used China's GLM 5.2 for forensics after US model guardrails blocked incident analysis, fueling a debate on cybersecurity restrictions.

A Russian-speaking hacker named bandcampro used Google's Gemini CLI to automate 89% of a criminal operation, including botnet control and server migration. The case reveals how AI is lowering the barrier for sophisticated cyberattacks.

Google is fixing an Android 16 bug that lets Gemini bypass lock screen authentication to send SMS and WhatsApp messages. The flaw, discovered by Bitdefender, affects multiple devices and raises serious security concerns about AI on the lock screen.

Sophos launches Fusion, an AI-native defense system built on open architecture to synchronize security controls. The platform aims to counter AI-powered ransomware and phishing with agentic autonomy and real-time response.