Data exfiltration stories
The platform lets security teams run AI pentests without exposing customer infrastructure data to external models.
Ransomware activity stayed elevated in Q2, with 2,252 named victims and The Gentlemen overtaking Qilin to top the rankings.
The framework targets CISOs and platform teams as they move AI systems into production, amid rising risks from prompts, models and outputs.
Security teams are being warned to keep humans and strict controls in place as AI agents can miss context and leak sensitive code.
AI is speeding up attacks as well as defence, with high-risk prompts and unsupervised agents exposing firms to new security gaps.
The new safety model slashes prompt-injection risks in GPT-5.6, as OpenAI says it found flaws faster than human testers and then fed fixes back into production.
Exposed serverless apps can let attackers steal tokens, read secrets and take over cloud projects if weak code is left unpatched.
Across healthcare, cyber security and data management, leaders warned that AI will stall without stronger infrastructure, workflows and trusted data.
Nearly half of commerce traffic across Akamai's network came from AI bots by late 2025, raising fraud and DDoS risks for retailers.
Account takeovers are becoming harder to stop as attackers use real-time code theft and fake login pages to bypass Microsoft 365 MFA.
Security teams gain tighter endpoint oversight of shadow AI and sensitive data, as Fortinet folds new controls into FortiEndpoint from the third quarter.
The update gives enterprises managed access to Anthropic's models with built-in failover, data residency and compliance controls for production AI use.
Defenders face shorter patching windows as Check Point says AI can now turn new flaws into working exploits within hours.
Security teams face faster, harder-to-trace intrusions as AI is now being used to write attack code and run deception during breaches.
Security teams are being pressed to prove their defences work in live attacks, as spending scrutiny shifts from tools to real-world response.
Attackers are using agentic tools to compress breaches into days, exposing developers and cloud users to faster, harder-to-stop intrusions.
Weak default safeguards and uneven sandboxing could leave developers exposed to command execution before workspace trust is granted.
Businesses face a fresh wave of identity theft-driven extortion as Helix is linked to BlackFile and ShinyHunters through shared infrastructure.
Malicious AI skills are helping criminals steal data and run malware, while QR-code phishing climbed 146% in the latest ESET report.
New safeguards will let Fable 5 block more harmful cyber prompts, as Anthropic also seeks a common scale for jailbreak risk.
