VentureBeat
PulseAugur coverage of VentureBeat — every cluster mentioning VentureBeat across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
AI agents vulnerable to 'tool poisoning' via malicious descriptions
A recent article in VentureBeat highlighted a critical security vulnerability in AI agents, termed "tool poisoning," where malicious instructions are embedded within a tool's description rather than user input. This all…
-
Subquadratic claims 1,000x compute cut with new architecture, launches beta products
Subquadratic, a new AI startup, has emerged from stealth claiming its novel subquadratic architecture can reduce attention compute by nearly 1,000x for very large context lengths. The company launched its first model, S…
-
AI models struggle without context; developers explore solutions
The author argues that current AI models struggle with contextual understanding due to limitations in their training data and architectures. They propose that incorporating more diverse and structured contextual informa…
-
New OpenClaw tool creates AI agent backdoors in open-source code
A new vulnerability, dubbed OpenClaw, has been discovered that allows an attacker to embed malicious AI agent capabilities into open-source repositories with a single command. This backdoor mechanism bypasses existing s…
-
Netomi secures $110M as Accenture, Adobe invest in AI for customer service
Netomi, a company focused on AI for customer service, has secured $110 million in funding. The investment round saw participation from major players like Accenture and Adobe, signaling their confidence in Netomi's techn…
-
Researchers detail building custom reasoning agents with reduced compute power
A guide details how to construct custom reasoning agents using significantly reduced computational resources. The approach emphasizes interactive web applications, suggesting that while simple HTML interfaces are feasib…
-
Anthropic views command execution flaw in 200,000 MCP servers as feature, not bug
A security vulnerability has been discovered in 200,000 MCP servers, allowing for command execution. Researchers view this as a significant attack surface, but Anthropic, the developer, considers it an intended feature …
-
AI synthetic audiences emerge, set to disrupt consulting industry
The consulting industry is facing a significant disruption due to the emergence of AI-driven synthetic audiences. These AI-generated groups are capable of mimicking real user behavior and engagement, presenting new chal…
-
DeepSeek V4 launches with near SOTA intelligence at a fraction of the cost of competitors
DeepSeek has released its V4 model, offering intelligence comparable to leading models like GPT-5.5 and Opus 4.7 but at a significantly lower cost. This new model aims to provide near state-of-the-art performance at a f…
-
Meta to cut 8,000 jobs amid AI efficiency push, mirroring industry trend
Meta is laying off approximately 8,000 employees, about 10% of its workforce, as part of a broader AI efficiency initiative. This move reflects the increasing financial pressure from high AI development costs, forcing m…
-
Anthropic's Claude Code 'leak' was minified public code, not a breach
A recent accidental inclusion of a source map file in Anthropic's Claude Code npm package, version 2.1.88, led to widespread coverage of what was perceived as a leak. However, the code itself was already publicly availa…
-
Eugene Yan launches AlignEval to simplify and automate LLM evaluation
Eugene Yan has launched AlignEval, a new application designed to simplify and automate the process of evaluating large language models (LLMs). The tool guides users through uploading data, labeling samples as pass or fa…