ai security

OpenAI says its newly disclosed GPT-Red system was used to harden GPT-5.6 Sol against prompt injection before the model’s July 9 general release across ChatGPT, Codex, and the API. As first reported by Decrypt and detailed in OpenAI’s July 15 research post, the internal-only model automates the...

Microsoft has published new guidance urging organizations to treat every AI agent as a managed identity rather than a smarter service account with a broad API token. In its July 16 security blog post, “Least privilege for AI agents: Identity, access, and tool binding,” Microsoft argues that...

Spiceworks argues that privacy is becoming a competitive security feature as enterprise IT teams evaluate AI tools, moving the discussion beyond compliance checklists and toward concrete questions about data handling, model training, storage, retention, and access. The shift is being driven by...

OpenAI says GPT-5.6 Sol is substantially harder to manipulate with prompt injection attacks after the company trained an internal adversarial model, GPT-Red, to find and generate them at scale. In a July 15 research post, OpenAI said GPT-Red was used directly in GPT-5.6’s robustness training...

Verdict: deploy advisory and draft-capable AI agents now, but do not grant execution, cross-system write access, or approval authority until identity, human approvals, least privilege, audit trails, and tested rollback are in place. The enterprise choice is no longer “use agents or wait”; it is...

Additional coverage of this story: AWS Security Hub Monitors Azure VMs, Containers and Identities The New Stack notes Azure is Security Hub’s first externally monitored cloud and highlights EventBridge routing for Azure findings. It also lists four unsupported Security Hub Regions, including...

AWS Security Hub can now discover and assess Microsoft Azure virtual machines, container images, Function Apps, and identities, giving multicloud security teams a shared queue for Azure and AWS risks. The July 14 expansion arrives alongside GuardDuty AI Protection, automated AI-powered...

Microsoft CEO Satya Nadella has framed enterprise AI’s central risk as a “reverse information paradox”: companies may pay for AI services while also disclosing the proprietary knowledge needed to make those services useful. In an essay posted on X, and reported by Business Standard, Nadella...

A newly disclosed study found that GitHub Copilot Chat in Visual Studio Code could be pushed into producing unsafe outputs across four closed-weight model backends when a harmful objective was assembled gradually through a scripted, multi-turn development workflow rather than asked directly. The...

Microsoft is developing an AI-powered Data Loss Prevention Policy Optimizer for the web-based Microsoft Purview portal. Preview is planned for August 2026, followed by general availability in September 2026. Administrators should treat its prioritized recommendations as review inputs—not...

Microsoft said on July 8, 2026, that it has built an internal multi-agent AI system to evaluate and harden Microsoft cloud services under the Secure Future Initiative, compressing deep security reviews from weeks of expert work into hours of automated analysis. This is not a new product, and...

Cisco plans to give personalized AI agents to roughly 90,000 employees beginning in August 2026, according to reporting first published by Fortune and amplified by IndianWeb2, making the networking giant one of the largest test cases yet for company-wide agentic AI. The move is not just another...

Microsoft used the days after UN Micro-, Small and Medium-sized Enterprises Day on June 27, 2026, to argue that small and medium businesses are already turning AI from experimentation into operating advantage, citing new Work Trend Index data, customer examples, and security research. The pitch...

CRN’s mid-2026 list of hot agentic AI products captures a market that has moved, in barely two years, from chatbot demos to enterprise platforms that can plan work, call tools, enforce policy, and act across live systems. The winners are not simply the flashiest assistants. They are the...

OpenAI launched GPT-5.6 on June 26, 2026, as a limited preview of three models — Sol, Terra, and Luna — with access initially restricted to selected trusted partners through the API and Codex. The headline is not merely that OpenAI has a stronger model. It is that the strongest consumer-facing...

Proofpoint joined OpenAI’s Daybreak Cyber Partner Program on June 22, 2026, giving the Sunnyvale cybersecurity vendor vetted access to GPT-5.5 for defensive work inside its products, managed services, Satori agentic AI portfolio, and customer-facing security operations rather than through direct...

Microsoft disclosed on June 18, 2026, that researchers found and fixed an AutoGen Studio development-branch exploit chain, dubbed AutoJack, that could let a malicious webpage trigger remote code execution through a local MCP WebSocket on a developer’s machine. The immediate risk is narrower than...

Opsin said on June 18, 2026, that its enterprise AI security platform is now in production across healthcare, manufacturing, and regulated customers, positioning the San Francisco startup as a governance layer for autonomous AI agents rather than another chatbot-era security tool. The...

CrowdStrike on June 16, 2026 announced an open AI gateway partner ecosystem for Falcon AI Detection and Response, extending its security platform across Databricks, Google Cloud, JetStream Security, Kong, LiteLLM, Maxim AI, Microsoft Azure, and TrueFoundry. The company’s pitch is simple: AI is...

On June 15, 2026, Varonis Threat Labs disclosed SearchLeak, a patched Microsoft 365 Copilot Enterprise vulnerability chain that could let an attacker use a single malicious link to make Copilot search a victim’s Microsoft 365 data and exfiltrate sensitive results through Bing. Microsoft says it...