Live CVE & 0day intelligence

Sound the depths of every vulnerability.

Newly released CVEs across every platform — sleek to read, verbose on demand, and served raw as markdown for AI and agent ingestion. Severity reads as depth: the deeper the contact, the graver the threat.

Browse the feed For agents → /llms.txt

Depth = severity + exploitation

HadalCritical and actively exploited (CISA KEV / 0day)
AbyssalCritical with a public exploit or in-the-wild use
MidnightCritical, or high with PoC / in-the-wild
TwilightHigh severity, or a signal on a lesser flaw
SunlitLow / medium · no exploitation signal

69227

Tracked CVEs

6415

Critical

701

Exploited in wild

4737

Last 7 days

Trending now

Bluesky + Hacker News mentions, last 7 days

1CVE-2026-109618Critical· 9.2RCE via OOM corruption4896
2CVE-2026-109619Critical· 9.2RCE via stale global state2196
3CVE-2026-111497Critical· 9.2Out-of-Bounds Write in Google Brotli Decoder1587
4CVE-2026-99999Critical· 9.3In the Linux kernel, the following vulnerability has been resolved: ipv4: validate checksum_start before completing checksum If a packet with bad checksum metadata gets into the ipv4 stack, skb_checksum_help can corrupt the network hea…963
5CVE-2026-110264Critical· 10.0Integer Overflow or Wraparound in GitLab267
6CVE-2026-108707Critical· 9.8Wukong_HRM through commit 186115e contains an authentication bypass vulnerability in ParamAspect that allows unauthenticated attackers to call every HRM API endpoint by omitting the AUTH-TOKEN header266

This week's digest →All digests

Feed

11 matches

CVE-2026-111160High· 7.5PoC

1w ago

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint. Attackers can send a sing…

▾ MidnightModelTC · LightLLMEPSS 0.42%via NVD

CVE-2026-108943Medium· 5.3PoC

2w ago

A weakness has been identified in ModelTC LightX2V up to 0.5.0

A weakness has been identified in ModelTC LightX2V up to 0.5.0. Affected is the function AutoTokenizer.from_pretrained of the file lightx2v/models/runners/cosmos3/cosmos3_runner.py of the component Tokenizer. This manipulation of the arg…

▾ TwilightModelTC · LightX2VEPSS 0.19%via NVD

CVE-2026-103270High· 7.5PoC

4w ago

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks. Unauthenticated attackers can call endpoints like /pause_generation, /abort_request, /flush_cache, and /init_weight…

▾ MidnightModelTC · LightLLMEPSS 0.51%via NVD

CVE-2026-103243Medium· 5.8PoC

4w ago

LightLLM through 1.2.0 fails to validate image_url and audio_url parameters in multimodal endpoints, allowing unauthenticated attackers to perform server-side request forgery

LightLLM through 1.2.0 fails to validate image_url and audio_url parameters in multimodal endpoints, allowing unauthenticated attackers to perform server-side request forgery. Attackers can supply arbitrary URLs to fetch internal resourc…

▾ TwilightModelTC · LightLLMEPSS 0.31%via NVD

CVE-2026-103395Critical· 9.8PoC

4w ago

LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method

LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method. Attackers can reach the visual RPyC port and…

▾ AbyssalModelTC · LightLLMEPSS 0.62%via NVD

CVE-2026-103042High· 7.5PoC

4w ago

LightLLM through 1.2.0 contains a memory exhaustion vulnerability in the NCCL control channel when started with --pd_trans_mode nccl, allowing unauthenticated attackers to exhaust KV-transfer worker memory

LightLLM through 1.2.0 contains a memory exhaustion vulnerability in the NCCL control channel when started with --pd_trans_mode nccl, allowing unauthenticated attackers to exhaust KV-transfer worker memory. Attackers can call the exposed…

▾ MidnightModelTC · LightLLMEPSS 0.65%via NVD

CVE-2026-103041Critical· 9.8PoC

4w ago

LightLLM through 1.2.0 multimodal deployments expose an unauthenticated RPyC cache service with pickle deserialization enabled on all interfaces

LightLLM through 1.2.0 multimodal deployments expose an unauthenticated RPyC cache service with pickle deserialization enabled on all interfaces. Attackers can send crafted serialized objects to exposed cache methods to execute arbitrary…

▾ AbyssalModelTC · LightLLMEPSS 0.77%via NVD

CVE-2026-103040Critical· 9.8PoC

4w ago

LightLLM through 1.2.0 contains a remote code execution vulnerability in the router profiler service when started with --enable_profiling flag

LightLLM through 1.2.0 contains a remote code execution vulnerability in the router profiler service when started with --enable_profiling flag. The service exposes an unauthenticated RPyC server with pickle deserialization enabled, allow…

▾ AbyssalModelTC · LightLLMEPSS 1.0%via NVD

CVE-2026-96560Critical· 9.8

1mo ago

LightLLM through 1.2.0 contains a remote code execution vulnerability in the KV-transfer worker when started with --pd_trans_mode nccl, which exposes an unauthenticated RPyC control channel that deserializes attacker-supplied data

LightLLM through 1.2.0 contains a remote code execution vulnerability in the KV-transfer worker when started with --pd_trans_mode nccl, which exposes an unauthenticated RPyC control channel that deserializes attacker-supplied data. Attac…

▾ MidnightModelTC · LightLLMEPSS 1.2%via NVD

CVE-2026-93839Critical· 9.8PoC

1mo ago

LightLLM through 1.2.0 contains an authentication bypass vulnerability in the /pd_register WebSocket endpoint that allows unauthenticated attackers to register arbitrary nodes by supplying crafted JSON without peer address validation

LightLLM through 1.2.0 contains an authentication bypass vulnerability in the /pd_register WebSocket endpoint that allows unauthenticated attackers to register arbitrary nodes by supplying crafted JSON without peer address validation. At…

▾ AbyssalModelTC · LightLLMEPSS 0.76%via NVD

CVE-2026-90919Critical· 9.8PoC

1mo ago

LightLLM through 1.2.0 contains a remote code execution vulnerability in the Config Server's unauthenticated /visual_register WebSocket endpoint that passes the first client frame directly to pickle.loads()

LightLLM through 1.2.0 contains a remote code execution vulnerability in the Config Server's unauthenticated /visual_register WebSocket endpoint that passes the first client frame directly to pickle.loads(). Attackers can reach the Confi…

▾ AbyssalModelTC · LightLLMEPSS 1.1%via NVD

From the blog

All posts →

Oct 27, 2026

Sound the depths of every vulnerability.

Trending now

Feed

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint

A weakness has been identified in ModelTC LightX2V up to 0.5.0

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks

LightLLM through 1.2.0 fails to validate image_url and audio_url parameters in multimodal endpoints, allowing unauthenticated attackers to perform server-side request forgery

LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method

LightLLM through 1.2.0 contains a memory exhaustion vulnerability in the NCCL control channel when started with --pd_trans_mode nccl, allowing unauthenticated attackers to exhaust KV-transfer worker memory

LightLLM through 1.2.0 multimodal deployments expose an unauthenticated RPyC cache service with pickle deserialization enabled on all interfaces

LightLLM through 1.2.0 contains a remote code execution vulnerability in the router profiler service when started with --enable_profiling flag

LightLLM through 1.2.0 contains a remote code execution vulnerability in the KV-transfer worker when started with --pd_trans_mode nccl, which exposes an unauthenticated RPyC control channel that deserializes attacker-supplied data

LightLLM through 1.2.0 contains an authentication bypass vulnerability in the /pd_register WebSocket endpoint that allows unauthenticated attackers to register arbitrary nodes by supplying crafted JSON without peer address validation

LightLLM through 1.2.0 contains a remote code execution vulnerability in the Config Server's unauthenticated /visual_register WebSocket endpoint that passes the first client frame directly to pickle.loads()

From the blog

VulnSea has moved to vulnsea.com (and what changed since September)

What's new: open corpus, three new MCP tools, vendor advisories, package pages

Wiring VulnSea into an agent: a hands-on tutorial

Sound the depths of every vulnerability.

Trending now

Feed

LightLLM through 1.2.0 contains a denial of service vulnerability that allows unauthenticated attackers to crash the inference worker by supplying the internal pd_kv_trans_params field to the /generate endpoint

A weakness has been identified in ModelTC LightX2V up to 0.5.0

LightLLM through 1.2.0 mounts reinforcement learning control routes on the public HTTP API without authentication checks

LightLLM through 1.2.0 fails to validate image_url and audio_url parameters in multimodal endpoints, allowing unauthenticated attackers to perform server-side request forgery

LightLLM through 1.2.0 visual_only deployments expose an unauthenticated RPyC service with allow_pickle enabled that deserializes attacker-supplied arguments in the remote_infer_images method

LightLLM through 1.2.0 contains a memory exhaustion vulnerability in the NCCL control channel when started with --pd_trans_mode nccl, allowing unauthenticated attackers to exhaust KV-transfer worker memory

LightLLM through 1.2.0 multimodal deployments expose an unauthenticated RPyC cache service with pickle deserialization enabled on all interfaces

LightLLM through 1.2.0 contains a remote code execution vulnerability in the router profiler service when started with --enable_profiling flag

LightLLM through 1.2.0 contains a remote code execution vulnerability in the KV-transfer worker when started with --pd_trans_mode nccl, which exposes an unauthenticated RPyC control channel that deserializes attacker-supplied data

LightLLM through 1.2.0 contains an authentication bypass vulnerability in the /pd_register WebSocket endpoint that allows unauthenticated attackers to register arbitrary nodes by supplying crafted JSON without peer address validation

LightLLM through 1.2.0 contains a remote code execution vulnerability in the Config Server's unauthenticated /visual_register WebSocket endpoint that passes the first client frame directly to pickle.loads()

From the blog

VulnSea has moved to vulnsea.com (and what changed since September)

What's new: open corpus, three new MCP tools, vendor advisories, package pages

Wiring VulnSea into an agent: a hands-on tutorial