AI Safety stories - Page 7

Nudge Security has launched AI agent discovery to help firms spot risky employee-built agents, tighten access and curb shadow AI sprawl.

Novee unveils an autonomous AI red teaming tool to probe LLM apps for prompt injection, jailbreaks and other emerging security flaws.

Orderfox readies Gieni ABX, an autonomous execution layer to turn its industrial AI insights into completed tasks across enterprise systems.

Gamma urges firms to phase AI customer experience rollouts, boosting governance and testing to cut disruption and data quality risks.

Sonatype says smaller AI tied to live software data can outsecure larger models on dependency upgrades, slashing risk and cost.

Barcelona startup Galtea raises USD $3.2 million to scale its AI agent testing platform and launch a self-service product for developers.

London mental health AI platform JAAQ raises GBP £13 million Series A to embed clinically governed support into employers' and insurers' apps.

NSS Labs warns many enterprise AI guardrails fail basic security tests, urging independent, real-world validation of protections.

SentinelOne unveils new AI security suite to govern autonomous agents, automate threat investigations and streamline data in its SIEM.

Yubico, IBM and Auth0 unveil an AI security model that forces human approval for high‑risk automated actions using hardware authentication.

BeyondTrust warns a surge of unsupervised AI agents is creating a hidden “shadow workforce” with admin-level access inside enterprises.

Oracle has woven new agentic AI tools into its core database and Fusion apps, promising secure, real-time automation on live enterprise data.

Ping unveils Identity for AI, a runtime identity framework to govern autonomous software agents' access, actions and accountability.

OpenAI sets out new safeguards for its Sora 2 video app, tightening controls on likeness, teens, harmful content and AI-generated audio.

Tanium rolls out AI-driven tools to tighten endpoint governance, speed incident response and unify security, exposure and device management.

AI-fuelled cyber attacks are spreading faster worldwide, CrowdStrike warns, as breakout times plummet and criminals weaponise mainstream tools.

Red Hat reports 97% of organisations suffered cloud-native security incidents last year, exposing basic failings in configuration and governance.

Upwind unveils sub-millisecond prompt threat detection for LLMs, claiming 95% precision using Nvidia models to secure live AI workloads.

Over half of enterprises now deploy generative AI, but an OpenText study warns security, governance and trust are lagging dangerously behind.

Most European organisations lack clear plans to rapidly shut down AI in a crisis, raising fresh concerns over governance and accountability.