Gemini’s new enterprise agent can delegate tasks, choose underlying models, and coordinate work across systems.

The company has disclosed another CVSS severity 10 flaw in its internet-facing appliance, just months after a separate pre-authentication flaw in the same product was exploited as a zero-day.

The 9.3-rated vulnerability lets unauthenticated attackers access specific files outside their intended boundaries. Atlassian is calling for immediate patching.

The vulnerabilities affect storage and server management infrastructure, with several enabling authentication bypasses, root access, credential theft, token forgery and remote code execution.

Employers can use AI to inform discipline and firing decisions, but a manager must corroborate the evidence, not merely blindly approve the machine’s recommendation.

Dots, Codex Cloud, and an expanded Agents API signal OpenAI’s push to make autonomous, long-running AI practical and cheaper for enterprise tasks.

The new Meta Enterprise Platform brings Muse agents, APIs, and coding tools together, but questions remain about Meta’s commitment to enterprises and its ability to earn their trust.

Google, OpenAI, and Anthropic are reportedly preparing to found a new industry body for frontier AI safety, but enterprises still need their own controls for model risk, data exposure, and agent behav...

Hacktron’s access to OpenAI’s internal monorepo shows why security teams need to distinguish what coding agents can read from what they can change or merge.

Two CVSS 9.8 flaws, including a newly disclosed zero-day in Check Point’s Security Management service, gave attackers a path into some of the most trusted systems on the enterprise perimeter.

CLOSEDQUORUM queries multiple models, tallies their decisions and automatically executes the winning action, eliminating the need for continuous attacker commands.

At least 45 US data center projects worth an estimated $68 billion were blocked or delayed in the second quarter, as communities increasingly push back on power, water, and land demands; analysts say ...

Architect Labs' AI system designed the Redwood inference chip in under two weeks. It runs on an FPGA, but its edge over Nvidia is still a silicon projection.

A new court order demands additional documents tied to allegations that Microsoft used licensing restrictions and discounts to push customers from perpetual software toward subscription services.

The incident shows why cloud modernization alone doesn't eliminate resilience risks, and why architects need to map dependency concentration and isolation and failure paths.

ECCO condemns VMware licensing changes and the shutdown of its main reseller program; SAP fares slightly better.

The vulnerability impacts self-managed CE and EE instances and provides an unauthenticated path to arbitrary file reads. It’s already being probed in the wild.

The ‘BlueMoon’ attack chains three high-severity Chrome and Windows flaws, raising concerns that sophisticated exploitation is becoming faster and more reusable.

A new framework explores how AI could affect GDP, unemployment, and wages, causing anything from modest gains to 15% annual GDP growth and nearly 1 in 5 cognitive workers unemployed.

The patch addresses two critical vulnerabilities and five other high-severity bugs impact Cisco’s carrier-grade routing OS, with potential paths to RCE and unauthorized access.

Thursday's outages across the three services highlight an uncomfortable reality: Enterprises are automating work faster than they're planning for AI failures.

The company is struggling to keep pace with the infrastructure demands of the agentic AI boom, with shortages now stretching across servers, storage, and other critical infrastructure components.

Google's 3.8 Flash Cyber model enhances vulnerability detection and patching, outperforming larger models at lower costs, crucial for cybersecurity teams.

Following three recent security incidents involving Claude, Anthropic is moving to more isolated environments, continuous monitoring, explicit instructions, and proactive human intervention.

The Tanzu Platform now combines AI-ready data infrastructure with sandboxed agents, isolated credentials, curated data products, and observability tools.

The vulnerabilities can be exploited remotely without user interaction; security teams should also look beyond ServiceNow to the credentials, APIs and workflows the platform can access.

Tempo’s new Workforce Intelligence offering ties AI activity and cost to individual Jira work items to measure its contribution alongside that of human labor.

Gartner has raised its 2026 semiconductor forecast, with revenue now expected to jump 92%. Memory alone will account for more than half of industry growth.

New VentureBeat Pulse Research finds enterprises run three AI orchestration platforms at once, yet 21% still can't stop runaway agent spending in real time.

Researchers found that adding LLM-generated rationales to recommendations can suppress productive human disagreement and cause evaluators to reject high-potential ideas.

MIT researchers found that as diffusion models scale, individual training examples have less measurable influence on outputs, and this “attribution decay” could complicate AI copyright, auditing, and ...

As agents reason, replan, call other agents, and work continuously in the background, Gartner predicts inference costs per workflow will rise more than fivefold through 2028.

DeepSeek's V4 Flash tops AI leaderboards but completed just 53.8% of real-world agent tasks in a new test — as the company hikes API prices up to 1,100%.

DeepSeek is raising its prices, to the dismay of developers, but off-peak pricing, cache discounts, and multi-model routing mean the impact is more nuanced.

Separate studies conducted by Google/MIT and Cloudera both point to the same problem: Agents can’t reliably act on data they can’t access, understand, or retrieve in real time.

The critical zero-day can provide direct SQL access to Metabase’s underlying database, potentially exposing credentials, API keys, and other sensitive data.

HTTP Terminator generated, evaluated, and weaponized thousands of HTTP desync hypotheses, and surfaced a new technique that neither the AI or its human operator would have found alone.

Cloudflare OS brings AI workspaces, zero-trust access, governed connectors, model routing, and enterprise data together in the browser, while new tools help IT teams track AI identities, token spendin...

Replit, Kilo Code, and Symbotic engineering leaders reveal how they track AI coding costs and stop runaway token spend before it wrecks the budget.

The vulnerability abuses Active Storage, allowing unauthenticated attackers to read sensitive files and potentially take over applications.

A study of nearly 150,000 real-world agent actions found AI increasingly handling low-risk IT tasks while human analysts iteratively shape and improve their actions.

The company plans to combine chat, agents and business workflows into a single AI workspace while separating memory, context and orchestration from any one foundation model.

At VB Transform 2026, Instacart CTO Anirban Kundu said AI-generated code gets rebuilt so often that tech debt has stopped being a concern for his team.

One startup said it cut cyberattack containment time from seven hours to twelve minutes. See four other approaches to running AI agents safely at scale.

A new guide from CISA and allied cybersecurity agencies says organizations should build dedicated isolation points into operational technology to contain attacks and continue delivering essential serv...

Target SVP Siobhán Mc Feeney says AI models aren't the company's real edge — the governance layer it built so every agent has to earn its autonomy is.

Lava researchers found more than 36,000 internet-exposed baseboard management controllers vulnerable to a 13-year-old IPMI flaw, giving attackers a path beneath the operating system.

By compromising captive Wi-Fi gateways instead of user devices, attackers can silently redirect authentication traffic and steal Microsoft 365 credentials.