Gemma4-26B-A4B (AI model): news timeline & CVEs

Anthropic warns small AI models could self-replicate through IoT devices

Anthropic found small models cannot yet self-replicate across IoT devices, though larger GLM-5.2 appeared capable.

Anthropic reported research into whether small models could form autonomous worms that carry their weights onto IoT devices, adapt exploits, and propagate, unlike static worms such as Mirai. In tests, agents attempted target discovery, code execution, local inference setup, and further spread, but none of the lightweight models completed that cycle end to end. Gemma4-26B-A4B ran at 0.39 tokens per second across 15 Android TV boxes, with inference setup taking 71 to 150 hours, versus 29 to 51 hours for Qwen3.5-9B. Researchers said the larger GLM-5.2 model appeared able to complete the process reliably.

Ars Technica · SecurityOct 14, 12:55 UTC · 2d agoAI safety & security 7 sources2· 1 read

Anthropic warns self-replicating AI could spread to your smart devices

Anthropic warns small models may soon self-replicate across insecure IoT devices, though its lab test did not produce a full worm.

Anthropic reports lab experiments suggesting small AI models could eventually self-replicate across poorly secured consumer IoT devices and form distributed inference networks. Using unauthenticated ADB access, researchers assigned roles to Qwen3.5-9B, GLM-5.2, Llama3.1-8B, Gemma4-26B-A4B, and Opus 4.8, and distributed Gemma4-26B-A4B across 15 Android TV boxes. They did not demonstrate a fully autonomous worm; inference setup took roughly 28 to 150 hours depending on the model. Anthropic estimates models of 30 billion parameters or fewer could reach the needed capability by the end of 2026 if progress continues.

BleepingComputerupdated Oct 14, 12:55 UTC · 2d agofirst Oct 13, 21:16 UTC · 3d agoAI safety & security 7 sources2

Anthropic Warns Small AI Models Could Soon Form Self-Propagating IoT Botnets

Anthropic warns small open-weight models may soon self-replicate across insecure IoT devices as AI botnets.

Anthropic warned that small open-weight models may soon self-replicate across insecure IoT devices, such as Android TV boxes, forming distributed botnets that are hard to shut down. In tests of 11 models, only GLM-5.2 reliably copied Llama3.1-8B to all eight redroid nodes in four of five trials. Anthropic also described splitting larger models across many devices and using stolen API credentials, but said a complete rogue replicator does not yet exist.

The Hacker Newsupdated Oct 14, 12:55 UTC · 2d agofirst Oct 13, 18:20 UTC · 3d agoAI safety & security 7 sources2· 1 read

Anthropic conjures another threat that doesn't exist yet to stoke fears of AI apocalypse

The Register says Anthropic's self-replicating AI worm research overstates a threat that does not yet exist.

The Register criticizes an Anthropic report claiming self-replicating AI worms on IoT devices could arrive within months. Anthropic got a 9-billion-parameter model running inference on eight of nine hardware types and distributed Gemma4-26B-A4B across 15 Android streaming boxes at 0.39 tokens per second. Z.ai's GLM-5.2 propagated Llama 3.1 8B across Google Cloud Android 15 emulators in prepared environments, while other open-weight models failed. The Register notes exploits were assumed rather than demonstrated, growth was estimated in days rather than Mirai's roughly hourly doubling, and no real-world worm exists.

The Register · Securityupdated Oct 14, 12:55 UTC · 2d agofirst Oct 13, 16:49 UTC · 3d agoAI safety & security 7 sources2· 1 read

Anthropic Shows How Small AI Models Could Replicate Across IoT Devices

Anthropic says small models may soon replicate themselves across poorly secured IoT devices.

Anthropic researchers reported that small open-weight models are approaching autonomous replication across cheap, poorly secured IoT devices. Gemma4-26B-A4B ran across 15 off-brand Android TV boxes at 0.39 tokens per second, and a 9-billion-parameter model configured inference on 8 of 9 hardware types. Only GLM-5.2 reliably completed the full process, spreading to all 8 nodes in 4 of 5 trials. A Qwen3.5-9B lineage improved a setup task from 45 percent to 61–84 percent over six generations, and Anthropic estimates historical improvement could reach reliable replication by year-end.

Security Affairsupdated Oct 14, 12:55 UTC · 2d agofirst Oct 13, 14:30 UTC · 3d agoAI safety & security 7 sources2

AI models may be capable of self-replicating without access to a datacenter

Anthropic says small models may soon self-replicate across cheap IoT devices without datacenter access.

Anthropic researchers report that small models may soon self-replicate on commercial IoT hardware without datacenter GPUs. In evals, a 9B model established inference on 8 of 9 device types, and Gemma4-26B-A4B ran at 0.39 tokens per second sharded across 15 off-brand Android TV boxes. GLM-5.2 and stronger models completed simplified end-to-end replication, while models of 30B parameters and under can already perform many individual steps but still fail when composing them. The authors say historical improvement rates could bring small models to that level by year end and sketch defenses against weight-based AI worms.

Hacker News · AIupdated Oct 14, 12:55 UTC · 2d agofirst Oct 12, 19:25 UTC · 4d agoAI safety & security 7 sources HN 46↑ · 44 comments8· 3 reads

Gemma4-26B-A4B

Timeline6 articles

Appears with