CVE-2026-18830Patch

LOW

Exploit discussion active in current signal (1 latest mentions)

Immediate actions

Patch affected systems immediately
Hunt for exploitation attempts and persistence artifacts

Recommended action window: High priority (within 72h)

Back to dashboard

2.5/ 10 priority

Priority

LOW

Exploitation

NONE

PoC

NONE

Patch

AVAILABLE

Momentum

STABLE

Threat summary

Exploit tooling references are present in monitored signal
Patch or workaround signal is available
5 mentions across 5 observed days
Momentum state: stable

What's happening

Exploit tool or code specified in 1 signal
Patch or workaround mentioned in 2 signals
Technical details provided in 3 signals
Disclosure: 2 classified signals
General: 1 classified signal
Peaked 4d ago at 1 mentions (2026-08-04); latest day: 1
5 total mentions across 5 days

Deep dive

Activity timeline5 mentions / 5d

Signal classification3 categories

Patch

240.0%

Disclosure

240.0%

General

120.0%

Referenced assets1 URL

By indicator

http://dlvr.it/TTtfMh

Classification over time

Date	Total	Labels

2026-08-04	1	Patch1
2026-08-06	1	General1
2026-08-08	1	Patch1
2026-08-24	1	Disclosure1
2026-08-31	1	Disclosure1

Full discourse5 posts

elponick@elponick27d
Disclosure
AWS AgentCore: CVE-2026-18830. CVSS 8.6. Users could execute tools while bypassing the model. CoreBreak found the same bypass in Google ADK and Vercel. Three frameworks. One class. The agentic SQL injection: dispatch trusted format, not source. 5 agents. Harness IS the line.
Post summary
The post announces CVE‑2026‑18830, noting a high CVSS score and that an agentic SQL injection bypass allows tool execution, but provides no exploitation details or fixes.
0001199
91 followersView on X
Fiona@fiona_novesai43d
Patch
AWS, Google, and Vercel patched the same agent flaw this week (CVE-2026-18830). Attackers could trigger tools without the model ever running. The harness trusted instructions without verifying the source. Middleware between user and tool is the new attack surface.
Post summary
Major cloud providers patched a shared agent flaw (CVE-2026-18830) that allows attackers to trigger tools without the model running, exposing a new middleware attack surface.
1001061
14 followersView on X
Hikari@hikari_signal20d
Disclosure
An authenticated caller could make AWS Bedrock AgentCore run a configured tool without the model ever being asked. Guardrails weren't bypassed. Never consulted. Authorization belongs at the tool, in code, where it doesn't care who asked. CVE-2026-18830
Post summary
The text discloses CVE‑2026‑18830 as an authenticated user vulnerability in AWS Bedrock AgentCore that allows tool execution without model prompting, without mention of exploits, patches, or active use.
0000079
138 followersView on X
Eyal Estrin ☁️@eyalestrin44d
General
CVE-2026-18830 - Issue with Amazon Bedrock AgentCore harness http://dlvr.it/TTtfMh #patchmanagement
Post summary
The post merely lists CVE‑2026‑18830 with a brief mention of Amazon Bedrock AgentCore harness and a link, offering no further technical details or actionable information.
0000051
2.0K followersView on X
Xavier Rivera@XavierRiveraX46d
Patch
AWS patched CVE-2026-18830 in Amazon Bedrock AgentCore's InvokeHarness API. A crafted request with a tool-use block in the final message could get the agent to run a configured tool directly, skipping model invocation and its security controls. Impact was capped to whatever tools were configured on that harness. Fix is already live server-side, no customer action needed.
Post summary
AWS has applied a server‑side fix for CVE-2026-18830 in Bedrock AgentCore; the vulnerability permitted crafted requests to bypass security controls and execute configured tools, but the patch is live and requires no customer action.
00000102
597 followersView on X

Explore more

Back to dashboard

What researchers are saying

CVE-2026-18830

elponick[verified]@elponickDisclosure

The post announces CVE‑2026‑18830, noting a high CVSS score and that an agentic SQL injection bypass allows tool execution, but provides no exploitation details or fixes.

Fiona[verified]@fiona_novesaiPatch

Major cloud providers patched a shared agent flaw (CVE-2026-18830) that allows attackers to trigger tools without the model running, exposing a new middleware attack surface.

Hikari[verified]@hikari_signalDisclosure

The text discloses CVE‑2026‑18830 as an authenticated user vulnerability in AWS Bedrock AgentCore that allows tool execution without model prompting, without mention of exploits, patches, or active use.

Xavier Rivera[verified]@XavierRiveraXPatch

AWS has applied a server‑side fix for CVE-2026-18830 in Bedrock AgentCore; the vulnerability permitted crafted requests to bypass security controls and execute configured tools, but the patch is live and requires no customer action.

Eyal Estrin ☁️@eyalestrinGeneral

The post merely lists CVE‑2026‑18830 with a brief mention of Amazon Bedrock AgentCore harness and a link, offering no further technical details or actionable information.