Technology is the core driver of Ant's future creation.
The ACM MM 2026 EgoLink Grand Challenge invites researchers to tackle egocentric video understanding by evaluating social reasoning, causal inference, intent prediction, and multimodal interaction, offering two tracks that test perception‑reasoning‑action loops on real‑world first‑person datasets.
The IIFAA Internet Trusted Certification Alliance, led by Ant Group and partners, has released the Agent Security Link (ASL) protocol, a verifiable security architecture that provides trustworthy identity, connection, authorization, and intent for AI agents, enabling secure, end‑to‑end interactions in high‑frequency scenarios such as parking payment and shared mobility.
Ling-2.6-flash is a 104B‑parameter Instruct model that uses a mixed‑linear architecture and token‑efficiency optimizations to achieve up to 340 tokens/s inference speed, 4× higher throughput than comparable models, and ten‑fold lower token consumption on Agent benchmarks, while maintaining SOTA performance.
This article reviews two recent ICLR 2026 papers—M²‑Miner, a multi‑agent Monte‑Carlo Tree Search framework for low‑cost mobile GUI data mining, and IGPO, an information‑gain‑based reinforcement‑learning method that provides dense rewards for multi‑turn search agents—detailing their designs, experiments, and open‑source releases.
The AT‑ADD competition, organized for ACM MM 2026, invites researchers to develop robust audio deepfake detection models across speech, environmental sounds, singing, and music, providing diverse real‑world datasets, baseline code, clear evaluation metrics, and a two‑stage submission process to advance AI security.
The article introduces YASA, a unified multi‑language static taint analysis framework built on a novel Unified Abstract Syntax Tree (UAST), explains its design, core components, open‑source releases, and benchmark results that demonstrate superior coverage, precision, and performance over existing single‑ and multi‑language tools.
Ant Group’s AI Security Lab and Tsinghua University have open‑sourced ClawAegis, a native security‑immune framework for OpenClaw agents that protects the entire lifecycle—from initialization to execution—by detecting malicious skill injections, memory poisoning, permission abuse, and providing dynamic auditing, configurable policies, and resource‑level safeguards.
A joint Tsinghua‑Ant Group study reveals a full‑lifecycle threat model for OpenClaw autonomous LLM agents, detailing five novel brain‑control attack vectors and proposing a five‑layer defense framework that secures the system from boot to execution.
A new Region‑to‑Image Distillation (R2I) approach lets multimodal large language models perceive tiny visual details in a single forward pass, eliminating costly tool calls while achieving state‑of‑the‑art accuracy on the ZoomBench fine‑grained benchmark.
Ling‑2.5‑1T is an open‑source instant large language model with 1 trillion total parameters, 63 B active weights, and a 1 M token context window, featuring mixed‑linear attention, a composite correctness‑plus‑process reward for token efficiency, fine‑grained alignment, and leading benchmark performance across reasoning, instruction‑following, and agentic tasks.
