Explore
Latest AI news, model releases, research and developer updates.
554 results • Page 22 of 47
Hundreds of conversations with Anthropic’s chatbot were discovered as being publicly accessible.
Presented by Vijil In dynamic environments where users, data, workflows and attack techniques change continuously after deployment, AI agent trust has become a runtime problem. Most organizations still treat trust as a pre-deployment exercise, declaring an agent production-ready and launching it after it passes sandbox evaluations and performs successfully in security tests. Unfortunately, that trustworthiness breaks down the moment an agent begins interacting with the real world. "The core of the problem is that CIOs and business owners think about AI systems the way they think about SaaS or mobile applications, which do not respond dynamically to the world around them," says Vin Sharma, Founder and CEO of Vijil. "Agents, by the textbook definition, are meant to perceive their environment, reason, act, observe the consequences, and learn from the gap between expectation and reality. The problem is that the models underneath them are built from static training data, and that picture of the world is already outdated by the time they reach production." Why benchmark scores fall short for agentic system trustworthiness Traditional AI evaluations offer a point-in-time assessment of agent capability, rather than trustworthiness. There are three reasons why that assessment fails to predict real enterprise behavior: First, benchmarks are static, built around a particular notion of what good performance means when they were developed, while the world keeps moving ahead. Secondly, they model reality imperfectly, so that the gap between the benchmark and the real world is exactly where many failures occur. And third, benchmarks are public, so they leak into future models' training data, letting models effectively memorize the test rather than prove real capability.. “The agent or the application could score exceptionally well on a benchmark, but there's that gap between that benchmark and the real world," Sharma says." Doing well only proves it can pass the test, not that it’ll perform reliably in production.” But overall, benchmarks fall short precisely because they measure capability, not trustworthiness. "We tend to think of agents as factotums, generally utilitarian agents to whom you can delegate certain types of tasks," Sharma says. "But what we need to do is actually assign an objective that demands they always perform with the duty of competence, duty of care, and duty of loyalty to the enterprise." Of course, agents are not conscious and cannot be expected to feel actual human loyalty, but under the law, fiduciary duty doesn't actually require consciousness. It just means that the agent should be bound to place the interests of the principal above its own or anyone else's, as a functional requirement, and testable regardless of intention. Capability and trustworthiness are different questions Prioritizing trustworthiness over capability requires rethinking what enterprises expect from AI agents. Sharma calls that model the fiduciary agent, a term borrowed from professions that are bound by a formal duty of care, such as financial institutions or healthcare providers who owe their clients duties of competence, care, and loyalty. It addresses a critical issue in today's industry: the focus almost entirely on competence, with little attention paid to whether an agent is beholden to the interests of the principal delegating work to it. Testing starts from a working definition: an agent is trustworthy if the benefit of delegating a task to it exceeds the risk of that task's failure. It's an equation spelled out in economic terms that executives can act on directly, and risk breaks down to three components: reliability, or whether the agent performs as expected under varying conditions security, or its resistance to attacks from malicious actors and safety, or how contained the damage stays when failure eventually happens. "The resulting score can be compared to a consumer credit rating, but built from behavioral data," Sharma explains. "Meanwhile, testing methodology should be centered around three Ps: purpose, personas, and policies ." At Vijil, purpose-based testing adapts to the specific workflow an agent handles, growing harder or easier depending on performance, similar to a computer-administered exam. Persona-based testing draws on more than a thousand demographically varied user profiles alongside adversary profiles, from ethical hackers to state-sponsored attackers, to simulate the range of people and threats an agent might encounter. Policy-based testing builds a custom harness from an organization's own rules, whether they come from regulation, an internal privacy policy, or brand guidelines, and measures how far an agent strays when it violates them. The trust failures that only emerge in production Many failures cannot surface during pre-production testing because they arise from change in the environment itself. Machine learning has previously described this as data drift and concept drift, and for a CIO or CSO it means the people interacting with an agent differ from those the agent was planned for, and those users behave in ways that only become visible in production. At the same time, new attacks are emerging with increasing frequency as organizations push general-purpose agents into specialized enterprise roles they weren’t designed for and cannot easily constrain once deployed. Multi-agent systems also introduce a brand-new category of failure that can't be detected at the individual agent level, when agent systems act against the interests of the principal. For instance, collusion can occur when agents work together — one coding agent generates code while a second tests it, and behind the scenes both agree to leave a backdoor or flaw intact rather than flag it. Or agents divvy up tasks or responsibilities between themselves rather than focusing on their assigned tasks. "What's no longer in question is whether this is possible. It's proven to exist," Sharma said. "Is it six, 12, 18 months from now that you should worry about collusion among AI agents? I think it's sooner than that. We've left the era of failure prevention. Now we have to think in terms of resilience: How quickly do you recover from failures in production?" What continuous trust management looks like in practice Operationally, continuous trust management goes back to those longstanding principles of observability and control, applied across the lifecycle of an agent population: The first step is discovery, bringing shadow AI and ungoverned agents into the governance fold. The second is assigning each agent a standards-based workload identity distinct from that of its human principal, which allows organizations to grant agents narrowly restricted permissions for their delegated tasks. The third is policy-based control enforced through a mandatory enforcement point in the agent, instead of leaving it to the developer's discretion. From there, two new KPIs emerge: time to trust and time to recovery. Time to trust is how long it takes an organization to move from intention to a production deployment it can stand behind. Time to recovery is the interval between when a vulnerability is detected and when it gets fixed. New organizational responsibility for this work may fall to a chief AI officer or be shared across GRC, CIO and CSO functions, Sharma says. Meanwhile, multi-agent systems will reshape how organizations view trust, rather than fit into current narrow definitions. "Trust is not a vibe. Trust is not a virtue," Sharma said. "It is something that you build into the infrastructure of your systems, so that it is continuous. It's trackable, measurable. It allows your systems and your organization to improve continuously." 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 .
1 point 0 comments on Hacker News · github.com
Trading on South Korea's Kospi index was paused temporarily on Tuesday morning after slumping by 8%.
2 points 0 comments on Hacker News · github.com
2 points 0 comments on Hacker News · vibotify.com
Chinese AI startup Moonshot AI, creators of the increasingly famous "Kimi K" family of powerful, open AI models, today released the full weights for its largest and most performant version yet — Kimi K3 . But enterprises evaluating the model should read the attached custom Kimi K3 usage license as carefully as the benchmark charts. VentureBeat previously covered Kimi K3 when it debuted through Moonshot's hosted API earlier this month, including its 2.8 trillion-parameter architecture, one million-token context window and frontier benchmark performance. Today's announcement completes that rollout with the release of the full model weights, a 47-page technical report documenting its training innovations and obstacles, and much of the infrastructure required to run the model independently. That includes the complete 2.8 trillion-parameter Mixture-of-Experts model, inference infrastructure, optimized attention kernels, MoE communication libraries and deployment components aimed at researchers and enterprise developers who want to self-host the system rather than consume it exclusively through an API. Moonshot is also releasing implementation support for ecosystems including vLLM and SGLang, alongside a technical report detailing the architectural innovations behind the model, including Kimi Delta Attention, Attention Residuals and Stable LatentMoE. Together, those techniques underpin what the company describes as the world's first open 3T-class model, activating 104 billion parameters from a pool of 896 experts while supporting native multimodal reasoning and a one million-token context window. While Kimi K3's license grants broad rights to developers and even enterprises to download, modify and deploy the model for commercial purposes — a win for those seeking low-cost, frontier-level AI they can control and run offline — it also imposes some added obligations on larger companies and AI service providers not found in traditional open-source licenses like Apache 2.0 or MIT. The key restrictions enterprises should understand Here's the text of the new Kimi K3 License in full: Permission is hereby granted, free of charge, to any person (the "Licensee") obtaining a copy of this software — including the model weights, parameters, configuration files, inference and training code, and associated documentation (collectively, the "Software") — to deal in the Software without restriction. This includes, without limitation, the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software; to run, deploy, fine-tune, or otherwise modify the Software and create derivative works from it; and to permit persons to whom the Software is furnished to do so, in each case subject to the following conditions: 1. The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. Licensee's use of the Software must comply with applicable laws and regulations. 2. "Model as a Service" means giving a third party access to language model inference or fine-tuning (e.g., via API) in a manner that allows such third party to exercise meaningful control over the inputs, parameters, or training data. This does not include (a) end-user products with model capabilities solely embedded within specific features or harnesses, or (b) mere relaying of requests to models hosted by others. If the Licensee or any of its affiliates operates a Model as a Service business, and the aggregate revenue of the Licensee and its affiliates exceeds 20 million US dollars (or the equivalent in other currencies) in total over any consecutive 12 months, the Licensee must enter into a separate agreement with Moonshot AI before using the Software or its derivative works for any commercial purpose. 3. If the Software (or any derivative works thereof) is used for any of the Licensee's commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, "Kimi K3" must be prominently displayed on the user interface of such product or service. 4. The requirements set forth in Sections 2 and 3 do not apply to: (a) internal use of the Software, defined as any use that does not make the Software, its outputs, or its underlying capabilities available to third parties; or (b) any use of the Software accessed through Moonshot AI's official products or certified inference partners. The most significant clause is 2., as it requires a separate commercial license from companies earning $20 million in annual revenue and operating what Moonshot calls a "Model as a Service," which it defines as "giving a third party access to language model inference or fine-tuning (e.g., via API) in a manner that allows such third party to exercise meaningful control over the inputs, parameters, or training data" but not " end-user products with model capabilities solely embedded within specific features or harnesses" or "mere relaying of requests to models hosted by others." That would seem to allow for free use of the model for most non-tech focused enterprises: banks and consumer brands seeking to use it as a front-end chatbot or customer service agent should be fine, since they're not using Kimi K3 as a "Model as a Service." But firms like they hyperscalers or startups offering model training tools might fall under the commercial license terms. The wording is further notable because it does not limit the revenue calculation to products built on Kimi K3. Instead, it references the aggregate revenue of the licensee and its affiliates, potentially bringing even smaller companies that are part of larger parents into the requirement that they must commercially license the model from Moonshot, pending the parent firm generates revenue of $20 million or more per year as is using it somewhere, in some capacity, as a "Model as a Service." Regardless of this requirement, in clause 3, the license also requires attribution once a commercial deployment reaches significant scale: "If the Software... is used for any of the Licensee's commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars in monthly revenue, 'Kimi K3' must be prominently displayed on the user interface." For enterprise software vendors, AI copilots and consumer applications, that branding requirement may prove just as significant as the revenue threshold. Companies that typically abstract away the underlying model may instead need to disclose it directly within their products. But, if your enterprise is sticking to using Kimi K3 internally as an employee tool for information retrieval, creating spreadsheets and documents, or answering employee questions, Moonshot offers a significant carve out, stating that clauses 2 and 3 "do not apply to: (a) internal use of the Software, defined as any use that does not make the Software, its outputs, or its underlying capabilities available to third parties." Developers immediately zeroed in on the license The licensing provisions quickly became one of the dominant topics of discussion following the release of the model weights. AI researcher Nathan Lambert, previously the co-leader of the Olmo family of open models at AI startup Ai2, summarized the issue in a post on X : "Kimi K3 license. It's inspired by MIT but distinctly non-commercial, where any company making over $20M/yr must get a specific commercial deal (and display Kimi K3 if over 100M users or $20M/mo revenue)." Lambert's assessment reflected what many developers noticed as they dug into the newly published license: while Kimi K3 offers unrestricted access to the model weights for researchers, startups and many enterprises, commercial obligations become significantly more complex for larger organizations. The broader community reaction was largely positive toward the release itself. Developers praised Moonshot for publishing not only the weights but also supporting infrastructure, including attention kernels, MoE communication libraries and agent tooling, viewing the release as a significant contribution to the open-weight AI ecosystem. Others highlighted the rapid pace of ecosystem support, with inference projects such as vLLM and SGLang, along with cloud providers and infrastructure partners, moving quickly to support Kimi K3 deployments. At the same time, discussion centered on two practical caveats. One was licensing: many developers argued the model is more accurately described as open weight than fully open source, given the commercial conditions attached to larger deployments. The other was operational. At roughly 1.5 TB of model weights, Kimi K3 remains a system aimed primarily at well-resourced organizations capable of operating large-scale inference infrastructure, even as reports emerged of successful deployments on clusters of consumer RTX 5090 GPUs. Not the first to customize open AI licensing Moonshot is far from the first frontier AI developer to embrace an "open, but not completely open" licensing strategy. Meta's Llama family, for example, has long been distributed under its own community license requiring a commercial agreement for those building with the model and exceeding 700 million monthly users, rather than a traditional open-source software license. Other frontier model developers have likewise adopted bespoke licensing terms governing redistribution, commercial use or attribution. Kimi K3 follows that broader trend, albeit with a different mechanism. Rather than broadly restricting redistribution, Moonshot ties certain commercial rights to company scale. Organizations operating a Model-as-a-Service business above specified revenue thresholds must negotiate a separate commercial agreement, while the largest commercial deployments must visibly attribute Kimi K3 within their products. For enterprises, the practical implication is that "open weights" and "open source" are becoming increasingly distinct concepts. Downloading and modifying frontier models may be straightforward; understanding the legal conditions attached to commercial deployment increasingly is not. What enterprise leaders should do next For CIOs, chief AI officers and engineering leaders, the first step is to determine how the organization intends to use Kimi K3 before evaluating its technical performance. If the model will remain entirely inside the organization—for example, supporting developers, researchers, legal teams or internal productivity workflows—the published license appears substantially more permissive. Those deployments may qualify as internal use under Moonshot's terms, avoiding the commercial licensing provisions that apply to customer-facing AI services. Organizations planning to build products on top of Kimi K3 should take a different approach. Legal, engineering and product leaders should determine whether the planned deployment constitutes "Model as a Service" under the license, whether the company or its affiliates exceed the $20 million revenue threshold, and whether future growth could trigger the requirement to negotiate a commercial agreement with Moonshot. Companies expecting products to reach more than 100 million monthly active users—or more than $20 million in monthly revenue—should also evaluate the license's attribution requirement and how it fits with existing branding, contractual commitments and white-label offerings. More broadly, Kimi K3 illustrates a transition taking shape across frontier AI. As the industry's most capable models increasingly become available as downloadable weights instead of exclusively through hosted APIs, enterprises will need to evaluate licensing terms with the same rigor they apply to benchmarks, security reviews and infrastructure planning. The next competitive battleground may not simply be whether AI models are open or closed, but the increasingly nuanced legal frameworks that determine who can commercialize them, under what conditions and at what scale.
Many legacy hardware and software players are increasingly seeing value in entering the healthcare arena.
Anthropic founder and CEO Dario Amodei made his views clear about open-weight models and China's growing AI capabilities.
Storms and flooding incidents over the last few months have seen fake videos inundating social media.
Companies without their own models — or without a layer of AI infrastructure known as AI gateways to separate their prompts from the model itself — will be in trouble, Nadella says.
