Red Teaming stories
Codex and ChatGPT users get a model that OpenAI says performs better on coding, research and office work while using fewer tokens.
As cyber tools become more powerful, Anthropic is limiting access while OpenAI is widening it, raising fresh fears over misuse.
AI-driven attacks are pushing firms to hide systems from the public internet rather than rely on patching flaws after discovery.
Rising AI-generated vulnerability reports are leaving security teams with record backlogs and only hours to judge which flaws hackers can exploit.
Boards are being pressed to abandon periodic patching as AI models can now uncover and chain software flaws faster than human teams can respond.
Financial regulators are alarmed after Anthropic said Claude Mythos can uncover software flaws at machine speed, raising bank security risks.
The framework is designed to expose hidden risks in production AI systems that can be missed by conventional one-off tests.
Security teams are bracing for harder-to-stop attacks after the model found a Linux kernel flaw that had gone unnoticed for 27 years.
AI tools are creating hidden east-west traffic that security teams struggle to monitor, raising the risk of data leakage and compromise.
The update promises better software engineering and longer task handling for users, while keeping Claude Opus 4.7 at the same price.
Brands using customer-facing chatbots face fresh pressure to prove safety and accuracy as Testlio rolls out human-led checks for live-use failures.
More than half of organisations have shipped AI tools, but quality problems and weak testing are leaving many projects stranded before production.
The hire signals Applause’s push into AI-driven testing as enterprises seek tighter checks on software before customer releases.
Offensive AI is widening exposure gaps for firms that test only a third of their attack surfaces on average, Synack says.
Boards in regulated sectors now have firmer assurance after Abacus secured CREST approval for penetration testing, renewed annually.
Security researchers say long automated jobs can make Claude Code’s deny rules fall back to user prompts, weakening protections in CI/CD pipelines.
Procurement teams in defence and critical infrastructure may now view White Rook Cyber more favourably after its CREST testing approval.
UK regulators are racing to assess whether Anthropic’s Mythos model could speed up attacks on banks and unsettle financial stability.
Many organisations overestimate their ability to recover from ransomware, as 57% of Irish respondents reported at least one attack in two years.
Enterprises face faster phishing, deepfakes and automated exploits as security leaders say existing controls lag behind frontier AI models.
