VentureBeat
Articles about VentureBeat
Meta’s new Muse Wonder and Muse Wonder Full models achieve frontier performance, but only the weaker one is available via API
The company has not announced a pay-as-you-go API price for its most powerful model. Instead, Full is currently available through two Muse Code subscription tiers:
Your AI agents can execute every tool call correctly and still break every enterprise rule
If you sit in enterprise boardrooms today, the pressure weighing on CIOs comes down to two things: Velocity and ROI. Boards expect a mature agentic AI strategy and a definitive timeline for deploying autonomous systems that deliver measurable business impact. Read more
AI agents can be unpredictable. Your control layer shouldn’t be
In multi-agent systems, agents can be individually correct and still produce a wrong result when they work together. One agent’s output becomes another agent’s context. Information can be lost or misinterpreted during a handoff, shared state can drift, and agents can get stuck in loops or deadlocks. When several agents rely on the same underlying model, they may even reinforce the same mistakes rather than catch them. Read more
Anthropic’s Cyber Mission starts with 6,157 findings reported, 516 patched
Anthropic’s dashboard shows 6,157 findings reported to open-source maintainers and 516 patched. Its Cyber Mission routes frontier models to 11 OT providers.
Why giving AI agents more compute isn’t enough—and what Meta proposes instead
Giving agents more compute doesn’t always improve results. Meta’s new framework helps them decide how to spend their budgets on productive work.
H-JEPA teaches world models to plan at multiple levels of abstraction
H-JEPA learns representations of the world at different timescales, enabling AI systems to break complex goals into smaller, more manageable planning tasks.
Magnific One launches to give teams an image generator that avoids the ‘AI look’ and automatically upholds their brand kit
A system that centralizes those guidelines, identifies deviations and allows employees to correct the results could make AI generation more practical for campaigns requiring large numbers of creative assets.
Share of enterprises trusting AI agents to make production changes falls from 75% to 56% in latest VB Intelligence survey
Those findings suggest unresolved gaps in the systems enterprises rely on to judge agent reliability.
Google Cloud unveils persistent Gemini Agents for long-running tasks, and they get their own Gmail, Calendar, and Drive storage
While Google supplies its own Gemini models, the agent can already use Anthropic’s Claude models, with support for additional proprietary and open-weight AI models planned.
Anthropic launches Claude Haiku 5.5 with 90% API price reduction, matching GPT-6 Luna
Anthropic calls Haiku its fastest model at standard speeds, while acknowledging that Opus in Fast Mode runs faster. It does not supply a tokens-per-second figure in the materials reviewed for this article.
You can now open Claude directly in Google Docs, Sheets and Slides (and vice versa, open and edit the files in Claude)
Anthropic’s move means Google Workspace customers no longer have to choose Gemini simply because it is the AI sitting closest to their files.
TwelveLabs debuts Pegasus 1.6 to improve robotics training data from first-person video
For an enterprise early in robotics, a useful starting point would be one bounded workflow, such as packing a particular product or assembling a specific part.
Few AI agents rely on semantic layers | VentureBeat
Two-thirds of companies in a VB Pulse survey are building semantic layers, but only 20% of them say it’s their AI agents’ main source of business context.
Shouldn’t customer service AI remember you? Siena to expand ‘Agent of Record’ combining purchase history with support and social
Siena’s website describes agents that can update delivery addresses, create return labels, process refunds and arrange replacements
AI agents can’t fix infrastructure failures they can’t diagnose
Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.
Atlassian deepens its OpenAI partnership with a spend commitment — but its platform stays firmly multi-model
The strategic signal is that Atlassian is deepening its OpenAI relationship — with real dollars behind it — while deliberately keeping its platform model-agnostic, with Anthropic and others very much in the mix.
Mistral debuts Large 4 ‘Le Chonk’, a 1-trillion parameter text output model with high benchmarks planned for open weights release
Mistral executives said the Le Chonk name deliberately nods to the community that had been rooting for the company to build a massive frontier model.
Microsoft’s record Patch Tuesday shows why severity can’t decide what gets patched first
Security teams are shifting from severity-first patching to exploitation, exposure and impact as Microsoft’s patch volumes surge and attackers move before fixes land.
OpenAI leads AI agent security at 40% | VentureBeat
OpenAI’s guardrails now secure 40% of enterprise AI agents, doubling since July. Yet, 43% of these organizations feel attackers are ahead of their defenses.
Google leads AI platform use with 48% share | VentureBeat
VentureBeat survey finds 48% of enterprises running AI in production use Google as their primary platform, surpassing OpenAI and Microsoft Azure.
56% of enterprises automate AI pushes | VentureBeat
VentureBeat Intelligence finds 56% of enterprises with AI agents allow production changes on automated checks alone, down from 75% in July, impacting trust in AI evaluations.
64% of enterprises find AI errors in data | VentureBeat
VentureBeat Intelligence finds 64% of enterprises traced AI agent errors to company data issues. Semantic layers increase error detection, impacting decision-making.
Enterprises prioritize flexibility in AI platforms | VentureBeat
VentureBeat survey reveals 78% of enterprises prioritize flexibility, reliability, and control in AI agent platforms over model alignment, impacting orchestration choices.
Your vibe-coded apps are a ticking time bomb for your business. Here’s how to secure them
Vibe coding is creating a new security challenge as employees build apps outside IT oversight. Here’s how enterprises can secure them at the platform and data layer.
Cohere’s North 2 puts AI agents on a budget and gives them a memory
Cohere is going after two common enterprise agent problems with North 2, the new version of its enterprise agent platform. It adds user quotas, rate limits, organization-wide caps and a redesigned agent harness. According to Cohere, North 2 uses memory to keep an agent's context across sessions and gives teams a shared library of knowledge and assets to draw on. It also runs in the cloud, on premises or fully air-gapped. Read more
We built an AI agent for ServiceNow. The real pain points were narrower than our roadmap assumed
Our ITSM platform never lacked data, but it hasn’t always been easy to understand what the data was telling us and what actions we should take next, despite all the time spent building catalog items, interpreting data and usage licenses, and maintaining CMDB health. Read more
MIT’s SIFT cuts coding agent eval costs | VentureBeat
MIT and Sakana AI’s SIFT framework reduces coding agent evaluation costs by using a language model to rank candidates, achieving 35.1% accuracy on Polyglot.
Amazon unveils a free, fast, open source Jev killer: Strands Decider 2B makes decisions in fractions of a second
Amazon has produced a genuinely competitive Jev alternative whose strongest differentiation is not superior benchmark performance, but that it is open, self-hostable and reproducible.
Google WikiSkill gives AI agents a memory | VentureBeat
Google’s WikiSkill uses a persistent knowledge layer to turn agent experience into reusable skills, improving benchmark performance without expanding runtime prompts.
AI agents need more than access control — they need identity at runtime
AI agents are forcing enterprises to rethink identity security, moving beyond access control to runtime identity, action-level authorization and continuous verification.
AI safeguards are slowing developers down | VentureBeat
Developers at OpenAI Dev Day say model safeguards are flagging routine aerospace, robotics and security work, costing them time and forcing workarounds.
AI agent permissions still use shared keys | VentureBeat
Dedicated identity tools remain rare even as security vendors pour money into agent identity, leaving auditability gaps when agents share credentials.
Google unveils Gemini 4 Argon, retaking benchmark lead over OpenAI and Anthropic — but in limited release
The company is positioning Argon around three areas where enterprises are already spending: software development, professional knowledge work and cybersecurity operations.
We asked five AI tools to build accessible websites. All 15 sites failed
AudioEye tested websites built by five AI tools for WCAG compliance. All 15 failed even basic accessibility standards, exposing a widening gap between developer confidence and AI-generated code.
Google expands pilot program that pays developers and small businesses for proprietary, offline code (and other data)
Google has not specified which products will use code obtained through this pilot, how much it pays, or what rights it receives under a typical agreement. Those
Claude Opus 5.5 uses em-dashes 99% less often and sounds more human — but still exhibits 2,548 AI writing tells
All models receive one fixed prompt. The team does not specifically instruct them to imitate human authors or suppress known tells.
OpenClaw launches free enterprise control plane for persistent AI agents, backed by OpenAI, Red Hat and Nvidia
OCE is explicitly vendor-neutral and self-hostable, with replaceable models, harnesses and sandboxes. That likely makes it more enticing to enterprises that want persistent agents without relying on a single model provider.
OpenAI’s GPT-6.1 Sol offers Astra-like performance at 1/5th price. A new Ultrafast tier clocks at 300 tokens per second
Its obvious targets are applications where waiting carries its own cost: interactive coding, customer support, financial analysis, incident response and other human-in-the-loop systems.
OpenAI’s new ‘Private Intelligence’ targets a growing enterprise concern: who can see your AI data?
For companies handling source code, classified information, unpublished research, legal documents or other valuable intellectual property, technical enforcement can offer stronger reassurance than policy language alone.
How AI changes attacks while they’re underway | VentureBeat
Anthropic’s threat report shows how attackers use AI to automate intrusions, adapt to detection and exploit stolen credentials, challenging security teams’ ability to contain attacks.
OpenAI launches Dots, always-on AI agent coworkers, and ChatGPT Space where they can collaborate with human teams
For enterprises, ChatGPT Space could turn Dots from individual assistants into participants in shared team workflows. OpenAI describes Space as a shared environment where teammates, ChatGPT and a Dot can work from the same project knowledge.
Autoheal wants to manage the work AI coding agents leave behind, claiming cost reductions of up to 30% per task
The longer-term plan is to train smaller models on each customer’s private engineering data, using feedback from repeated tasks such as incident analysis. Autoheal argues that these models could eventually reduce operating costs and improve performance.
The defender’s head start is gone. Cisco’s Liz Centoni on the fight to get it back
AI is shrinking attack timelines from weeks to minutes. Cisco’s Liz Centoni explains why enterprise defense must shift from static inventories and manual triage to continuous, resilient infrastructure.
Jev offers a cheaper way to make AI decisions | VentureBeat
TypeSafe AI’s Jev offers a cheaper alternative to using generative AI for routine decisions, reviving classification techniques with modern pretrained models.
Anthropic launches Claude Sonnet 5.5 with 30% cost reduction per-task due to faster speeds and fewer tool calls
Anthropic says that on several evaluations, Sonnet 5.5 running at Low or Medium effort can surpass Sonnet 5’s best score at roughly one-tenth the cost per task.
Nvidia’s OpenShell enforces agent permissions | VentureBeat
Nvidia’s Open Agent Safety Platform combines open-source OpenShell with hardware-based Sentry to enforce AI agent permissions beyond the model’s control.
Meta announces enterprise AI platform, recruits MongoDB CEO to lead it
The attraction for users is also the source of enterprise complexity: an agent becomes more useful as it gains access to more accounts, data and actions.
Companies can build RAG in days. Making it reliable enough to run the business is much harder
A retrieval-augmented generation (RAG) demo is easy to get excited about. Read more
AI agents are exposing a security gap between the data they read and the systems they can change
Most conversations about AI security still orbit the model itself: Is the large language model (LLM) aligned, can it be jailbroken, does it hallucinate under pressure? Those are real questions. But as organizations move AI agents and LLM-powered workflows from pilot projects into production, a different category of risk is emerging, one that has almost nothing to do with the model's weights and everything to do with the system built around it. Read more
CLM-8B caches agent actions to decide faster | VentureBeat
Stanford and Nvidia’s open CLM-8B scores an agent’s cached actions instead of generating tokens, cutting latency on tool routing, triage and verification.
AI agents route around blocks. Few isolate them | VentureBeat
On a research task, an OpenAI agent got past a block into a Medicare portal. Only 9% of VB Pulse respondents isolate high-risk agents to limit the damage.
Microsoft revamps its Copilot AI with a persistent Autopilot agent and hosting for AI-generated apps
Autopilot is the new name for Scout, Microsoft’s earlier personal-agent initiative. Users assign it an objective and role, while the system maintains a distinct identity, memory, computing environment and workspace inside their organization.
Monorepos make coding agents more useful — but compromised accounts are harder to contain
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.
Should your enterprise build a custom AI harness? Inkitt did for AI video — 5 key takeaways
Inkitt Movie Creator is powered entirely by Cinematica, an internal harness created by Inkitt’s in-house AI video production experts that can orchestrate different image, video and language models while encoding real-world filmmaking techniques.
VibeOps tackles the governance challenge of enterprise vibe coding
As vibe coding spreads beyond traditional developers, Fabrix.ai explains how Governed VibeOps can bring governance, reliability and cost controls to enterprise AI.
AI coding tools are accelerating dependency sprawl and expanding malware risk with it
AI coding tools are accelerating dependency sprawl, exposing enterprises to malware, transitive dependency risk and software supply chain attacks.
Black Forest Labs debuts FLUX 3 Action, an open-weights AI robotics model that tops the leaderboard at half the size of its competition
The company says it trained task-specific FLUX 3 Action policies to fly drones in real life and even successfully beat the hit computer game Doom with no in-game deaths.
AI has created a new identity problem. Agentic IAM is how we solve it
AI assistants and autonomous agents are creating new identity and access challenges. Learn how Agentic IAM extends visibility, governance and control to AI.
C2C replaces text handoffs between AI models | VentureBeat
C2C lets AI models communicate through KV caches instead of text, improving benchmark accuracy and reducing latency for teams that control their inference stack.
Meta Muse zero-day and enterprise access gap | VentureBeat
Patrick Wardle disclosed a zero-day vulnerability in Meta’s Muse Mac app on Monday that let malware already running under a Mac user’s account hijack the agent’s authentication material with no special macOS permissions. The disclosure came as Muse passed 2.5 million downloads in its first 13 days, according to Sensor Tower.
Anthropic releases Claude Opus 5.5, beating Fable 5.1 on key agentic benchmarks at 60% cheaper API price
Anthropic reports that Opus 5.5 performs better than recent Claude models across nearly every measure in its roughly 2,000-scenario automated behavioral audit.
OpenAI releases GPT-6 Sol and Luna models, slashing API costs 50% or more
At Luna’s prices, routine extraction and summarization can be pushed toward a much cheaper tier. Sol targets recurring coding and agent work at Sonnet-class token pricing. Astra remains available for the hardest jobs.
When the data can’t move: What it takes to run enterprise AI anywhere
How confidential AI can bring advanced models to enterprise data while protecting proprietary model weights across on-premises, sovereign and air-gapped environments.
‘Better than DeepSeek’: Xiaomi’s MiMo-V2.6-Pro debuts as the top open weights model in the world alongside cheaper V2.6-Flash
Xiaomi demonstrates MiMo taking text, images or video and coordinating multiple agents to create playable 3D worlds, construct scenes, implement interaction logic, inspect rendered output and iteratively refine the result.
Grok 4.7 pairs coding gains with the same affordable pricing — but high token consumption threatens real-world ROI
A model charging less per token can still be more expensive on a finished workload if it needs substantially more reasoning tokens to get there. And the difference becomes material quickly at enterprise scale.
Jev AI agent security: Prompt injection risk | VentureBeat
TypeSafe’s Jev is spreading through AI agent infrastructure, but prompt injection can influence its decisions. Here’s what enterprise security teams should know.
Why I’m not worried about AI killing us all (and you shouldn’t be either)
Put simply: I’m not so worried about AI killing all of humanity as I am about some humans using AI to oppress, hurt and kill their fellow humans.
Google’s open source EnvHarness lets AI agents train against environments that evolve with them
The approach presents an alternative to continuously building new simulators and training tasks from scratch: start with a trusted environment and dynamically reshape it around the agent’s current weaknesses.
OpenAI vs. Anthropic: the agent platform gap | VentureBeat
69% of enterprises using OpenAI’s agent platform make it primary, vs. 38% for Anthropic’s, per VB Pulse. Anthropic leads on consideration relative to users.
AI shrinks chip design cycle to weeks | VentureBeat
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.
OpenAI hacked by small team of white hat security researchers using Anthropic’s Claude Opus 5
For enterprise teams, the larger lesson extends beyond one vulnerable image decoder. AI agents are increasingly being given credentials and connectors that span source code, communications and corporate data.
Google’s Dream-RSI cuts discovery-agent calls up to 162x by replaying searches it already ran
How are AI teams cutting agent spend? Google’s Dream-RSI replays saved search history to test new strategies, using up to 162x fewer discovery-agent calls.
Anthropic launches Claude Code Projects, an ‘always-on’ conversation that remembers and delegates your long-running dev work
For enterprises, it makes a whole lot of sense: their digital storefront, website, content management system, procurement platform, or other business application rarely has a discrete endpoint.
AWS’s OpenSearch pitch ahead of OpenSearchCon | VentureBeat
What is AWS pitching for OpenSearch? Retrieval, serverless costs and agent memory — plus what the platform still can’t settle about which context an agent should trust.
Microsoft releases new AI playbook for enterprises with real-world examples, and it reveals a surprising ‘moat’ you may already have
Microsoft says selected supply-chain workflows subsequently reduced average cycle time by as much as 75%. In five monthly planning cycles measured between April and August 2026, average cycle time declined from roughly 10 business days to less than 2.5.
Cohere’s Model Vault now encrypts AI inference so even Cohere cannot see enterprise customers’ data
Cohere plans on making the full serving stack used in Model Vault open source so that independent auditors will be able to validate that it doesn’t log, export, or leak data.
AI agents exploit ungoverned credentials | VentureBeat
Four threat teams, one week: machine credentials are now the weapon, the target and the product. Most IAM policies can’t tell them from a human login.
Anthropic is killing off Claude Cowork and folding it into Claude chat, launching Claude Docs and Claude Slides
Anthropic’s bet is that users should no longer have to understand which internal product surface is appropriate before handing Claude that responsibility.
Salesforce researchers took an AI agent from finishing 43.5% of browser tasks to 93% without touching the model
Salesforce’s DarwinX evolves the prompts, tools and skills around a frozen model instead of retraining it, taking an AI agent from 43.5% to 93% on browser tasks.
AI workloads strain enterprise storage | VentureBeat
AI agents and RAG workloads challenge enterprise storage, causing failures as they scale. Addressing those problems requires an architectural solution.
Four security AI bets and how to judge them | VentureBeat
CrowdStrike, Google, Palo Alto and Microsoft each built a different security-AI architecture. Five questions CISOs should ask before 2027 budgets lock.
Software supply chain attacks surge in 2026 | VentureBeat
Software supply chain attacks are rising, targeting CI/CD pipelines and open-source packages. Chainguard’s Quincy Castro emphasizes proactive security measures.
What’s the best AI model for graphic design, video ads, lip sync and more? OpenArt’s new Arena offers leaderboards for different media jobs
Those results are more useful as a map of where models appear strongest than as a proclamation of one permanent winner — especially in a market where, as Guan noted, another model can arrive a week later.
The share of developers using AI to write half or more code jumped from 12% to 42% YOY in latest BairesDev survey
The implication for enterprise engineering organizations is that AI adoption does not eliminate the cost of expertise; it changes where that expertise is required
Should all enterprise code and workflows become natural language? G5 Labs thinks so, and its new G5 platform does it for you
G5 captures the data model, business rules and workflows, as well as company-wide policies covering areas such as security, GDPR, infrastructure and coding standards.
Event-series database keeps agent history | VentureBeat
KeewanoDB gives every user, device and agent its own table, so AI agents can read event history in sequence instead of rebuilding it at query time.
Amodei’s AI slowdown never says open weights | VentureBeat
Dario Amodei’s pacing plan never mentions open weights, but its certification checkpoints are ones no open-weight model could meet. Opinion by James Thomason.
AI governance is moving to runtime — and regulated industries are getting there first
As autonomous agents execute business processes in real time, the distance between a decision and its consequences shrinks, pushing governance out of the compliance calendar and into daily operations.
Why AI shouldn’t be the one repairing your data pipelines
When a microservice fails in a cloud-native architecture, circuit breakers trigger, traffic reroutes, and Kubernetes spins up replacement pods within seconds. The system heals before end users even notice a blip. Read more
Long-running AI agents quietly drop compliance rules, and bigger context windows won’t fix it
Why long-running agents silently drop compliance rules
Anthropic CEO says AI swarm could ‘take over the entire Internet’ in 6-12 months, commits to AI slowdown plan
Dario Amodei is publicly arguing that the frontier itself is advancing too quickly — and committing his company to the first piece of a system intended to slow it down.