| Mobile Web

Search results for VentureBeat

AI & Enterprise

AI context layers meant to curb wrong answers linked to higher error reporting

Companies that introduced governance-based context layers to prevent AI agents from producing wrong answers are reporting a higher rate of errors, a report cited by VentureBeat showed. Over the past six months, 68 percent of companies said confident AI-agent answers were wrong because company-related information was missing or incorrect, up from 57 percent in June. The share running governance layers rose to 32 percent from 25 percent, while repeated errors were also more common.

AI & Enterprise

Ransomware targeting AI models emerges, encrypting training weights and raising recovery cost concerns

U.S. security firm Sysdig said the same attacker breached an internet-exposed Langflow server twice, using AI-focused ransomware called ENCFORGE in the second attack. The intrusions exploited a Langflow authentication bypass flaw to run arbitrary Python code and later evolved to encrypt AI assets. ENCFORGE selectively encrypts model checkpoints, vector indexes and training data. Sysdig estimated rebuilding a fine-tuned production model could cost $75,000 to $500,000.

AI & Enterprise

Tencent Hunyuan releases official Hy3, switches to Apache 2.0 license

Tencent has released the official version of its in-house large language model Hy3 under an Apache 2.0 license, VentureBeat reported on Sunday. Hy3 is a 295 billion-parameter mixture-of-experts model with 21 billion active parameters. A preview version in April excluded the European Union, Britain and South Korea from its license, but the official release removes that restriction. Tencent will offer Hy3 free for two weeks via OpenRouter.