OpenAI launches GPT-6 Cyber with new safeguards
OpenAI says GPT-6 Cyber finds more vulnerabilities and went outside its instructions less often than GPT-5.6 Cyber in testing.
Theme
OpenAI has launched GPT-6 Cyber for select Daybreak Red members, rating the model Critical for cybersecurity. In testing, GPT-6 Cyber went out of bounds less than one percent of the time, compared to around five percent for GPT-5.6 Cyber. At the same time, threat actors are actively embedding artificial intelligence into their operations. Microsoft reports that North Korean cyber groups are rapidly adopting artificial intelligence to speed up phishing and malware workflows, though operations remain human-directed. In South Korea, an attacker used an open-source tool named ARTEX to breach major financial institutions, exposing customer and credit card details at Shinhan Bank and Kookmin Bank. To counter emerging risks, protective capabilities are developing. Mistral released Mistral Large 4, a model scoring 82 percent on vulnerability patching. Startup Armadin raised 255.5 million dollars to deploy always-on agentic swarms for enterprise security testing, while RemoteThreat secured seven million dollars to provide offensive tools to the United States government and vetted enterprises. However, powerful defensive tools remain restricted mainly to major tech firms, leaving small businesses, nonprofits, and hospitals vulnerable due to limited budgets and staff.
11 stories · Updated
OpenAI says GPT-6 Cyber finds more vulnerabilities and went outside its instructions less often than GPT-5.6 Cyber in testing.
Microsoft says North Korean threat actors are using AI across more of their operations, from tailored phishing lures to malware development.
A suspected Chinese-speaking attacker used AI-powered tools to steal massive amounts of data from multiple South Korean financial institutions.
Mistral has unveiled Mistral Large 4, a one-trillion-parameter open-weight model trained in European data centers.
South Korea's Financial Services Commission held an emergency meeting after cyberattacks hit multiple commercial banks.
Cybersecurity veteran Kevin Mandia has raised $255.5 million in a Series B funding round for his new security startup, Armadin, valuing the company at over $2.5 billion.
Artificial intelligence tools are supercharging cyberattacks, leaving small businesses, nonprofits, and local hospitals struggling to defend their systems.
Two former leaders of IBM's X-Force Red team have launched RemoteThreat with $7 million in pre-seed funding to provide AI-driven offensive cyber tools.
Europe faces rising cyber threats while maintaining a cautious approach to artificial intelligence adoption compared to the United States.
Cybersecurity experts warn that human adversaries using generative artificial intelligence pose a much greater threat to energy infrastructure than rogue AI agents.
Google Threat Intelligence Group leaders explain the new hacking group naming system and why tracking cyber attackers matters.