Red Teaming stories
Verified access to Anthropic's restricted AI tools could help IRONSCALES test email defences against more realistic phishing and impersonation attacks.
The public test could bolster or undermine claims that VEIL can anonymise sensitive AI data without letting outsiders recover the original records.
AI systems and social engineering tests proved especially risky, as CyberCX found severe weaknesses in half and 77% of cases respectively.
Security teams face a broader threat as criminals and state-backed actors use generative AI to speed hacks, phishing and malware.
CurricuLLM rolls out a school AI monitoring tool in Australia and New Zealand, flagging 21 harm types from academic offloading to personal revelations.
Vetted security teams will get fewer refusals on authorised tasks as OpenAI tightens access around its most permissive cyber model.
The tie-up could help security teams cut false alarms and patch faster as automated attacks shrink defenders’ reaction time.
The move aims to widen security coverage as firms struggle to test expanding attack surfaces quickly enough.
Enterprises using Microsoft Defender will get round-the-clock human-led threat hunting, as CrowdStrike also broadens its AI risk coalition across partners.
Many security teams are deploying AI before proving it works, with readiness scores as low as 30% despite 78% confidence.
Detection of malicious code can collapse when AI reviewers are fed large files packed with harmless text, Cloudflare's research shows.
No patch exists for a Windows RPC weakness that can let attackers turn a service foothold into SYSTEM-level control on a host.
Businesses could face faster cyber attacks as experts warn Anthropic's leaked Mythos model may outpace remediation and widen governance gaps.
Businesses face tighter reporting and new rules as ministers move to overhaul cyber security, AI oversight and digital identity regulation.
Vulnerability exploitation has collapsed from years to hours, leaving organisations racing to fix exposed systems before attackers do.
Banks and fintechs are being pushed to sharpen cyber defences as AI threats and operational knock-on effects test the UK payments system.
The findings add pressure on ministers to modernise the 1990 Computer Misuse Act as breaches hit 43% of UK businesses and 28% of charities.
Seven critical weaknesses were found in live production systems over a weekend, showing AI-driven pentests can now uncover basic flaws cheaply.
Repeated phishing training helped cut Singapore staff click rates to 7.4% from 17%, despite more than 8,500 fake emails sent.
Defenders face faster, harder-to-stop attacks as SANS says AI is now built into phishing, malware and reconnaissance at scale.
