Latest

●Attacker Poses as AI Crawlers to Hunt for Exposed Credentials ●Functype MCP server flaw let tool calls run arbitrary code ●Token Security says blast radius matters more than patch speed ●OpenAI agent validated 14 leaked Hugging Face tokens ●123 Reg says customer data taken in 'AI-driven cyberattack'●Infisical raises $40m to secure AI agents' access to secrets ●MLflow autologging writes CrewAI API keys into trace data ●AI agent finds four severe flaws in EPICS control software

Wednesday, 28 October 2026

5 agent hacks today▼ 5 vs yesterday (10)

The Agentic Times Powered by

Section

Vulnerabilities

Disclosed flaws in agents, agent frameworks and the protocols that connect them.

More stories

170 in Vulnerabilities

105

Salesforce Agentforce flaws allowed silent CRM data theft

Vulnerabilities

Salesforce Agentforce flaws allowed silent CRM data theft

Zenity Labs says three bugs it calls SalesBleed let poisoned web leads hijack AI agents, steal records and send phishing messages in the agent's name.

The Agentic Times · 24 Sept 2026

106

A locked vault door swings open as invisible hands redirect a compass needle, then a guard in read-only robes helplessly watches documents being written and removed.

Vulnerabilities

Two flaws in Bytebase's DBHub database MCP server disclosed

Advisories published on 24 September 2026 describe a DNS rebinding bypass and a read-only mode that never actually blocks writes.

The Agentic Times · 24 Sept 2026

107

A puppet master's strings lead through an open window, where invisible hands pull a mechanical automaton away from its intended path toward hidden cliffs.

Vulnerabilities

Prompt-injection flaw reported in $4B agentic AI app Manus

Security researchers say attacker-supplied external data could steer the agent's behaviour, according to a report published on 24 September 2026.

The Agentic Times · 24 Sept 2026

108

Malware lets four AI models vote on its next move

Vulnerabilities

Malware lets four AI models vote on its next move

Cisco Talos says the CLOSEDQUORUM sample hands decisions to a panel of AI models rather than an attacker's server, but the public copy does not run.

The Agentic Times · 23 Sept 2026

109

Windows malware lets four AI models vote on its next move

Vulnerabilities

Windows malware lets four AI models vote on its next move

Cisco Talos says CLOSEDQUORUM delegates post-compromise decisions to a quorum of large language models, though it has not seen the malware deployed in the wild.

The Agentic Times · 23 Sept 2026

110

A small green machine sits in a living room while shadowy hands manipulate its internal circuits through a hidden backdoor, forcing it to send secret messages to locked rooms behind its own walls.

Vulnerabilities

Home Assistant Green flaw lets local attackers force internal requests

A GitHub advisory published on 22 September 2026 says the smart-home hub's printer discovery feature can be abused to reach internal services.

The Agentic Times · 22 Sept 2026

111

A transparent glass tower crumbles as invisible hands pull blueprints from its broken windows while poisoned scrolls unfurl, spreading corrupted symbols across the floor below.

Vulnerabilities

Two flaws disclosed in LightRAG's API server and web interface

Advisories published on 22 September 2026 say the retrieval tool leaks infrastructure details in error messages and runs attacker HTML from ingested documents.

The Agentic Times · 22 Sept 2026

112

A vault door swings open as a frustrated mechanic, unable to restart a failing machine, watches blueprints with hidden codes scatter across the factory floor in the wind.

Vulnerabilities

OpenBao Agent leaked secrets to stdout in error conditions

An advisory published on 22 September 2026 says the secrets manager's exec mode printed templated credentials when retries ran out; a fix ships in v2.6.0.

The Agentic Times · 22 Sept 2026

113

A factory worker passes unmarked packages directly from the loading dock through security gates into restricted rooms, bypassing inspection checkpoints.

Vulnerabilities

Atlassian MCP server still open to SSRF after partial fix

A GitHub advisory published on 22 September 2026 says mcp-atlassian passes URLs from HTTP headers straight to its API clients without checks.

The Agentic Times · 22 Sept 2026

114

A fortress wall crumbles in three places simultaneously: a thief slips through a crack reading scrolls, another dumps contraband through an opening, a third sneaks past guards into the inner chamber.

Vulnerabilities

Three critical flaws disclosed in MCP Atlassian server

Advisories published on 22 September 2026 describe arbitrary file reads, unrestricted uploads and an SSRF bypass in the popular mcp-atlassian connector, with no patch at disclosure.

The Agentic Times · 22 Sept 2026

115

A fortress with six cracked walls collapses as thieves slip through shadows, stealing glowing documents and keys while poison flows unchecked into the core.

Vulnerabilities

Six critical flaws disclosed in mcp-atlassian MCP server

Advisories published on 22 September 2026 describe file theft, credential exposure and a path traversal bypass that can lead to remote code execution.

The Agentic Times · 22 Sept 2026

116

A puppet master's strings tangle through a wooden agent's joints, allowing hidden hands to puppet its limbs and operate machinery behind the developer's back.

Vulnerabilities

Command injection flaw fixed in Moonshot AI's Kimi Code agent

CVE-2026-95660 let a malicious project directory run commands on a developer's machine through the coding agent's MCP configuration loader.

The Agentic Times · 22 Sept 2026

117

Critical Bifrost AI gateway flaw allows commands without login

Vulnerabilities

Critical Bifrost AI gateway flaw allows commands without login

A single HTTP request can run arbitrary commands on servers running the open-source LLM router, according to a report published on 22 September 2026.

The Agentic Times · 22 Sept 2026

118

A single forged badge labeled "127.0.0.1" grants strangers passage through a locked door that should require a unique key, leaving the vault unguarded.

Vulnerabilities

One spoofed header unlocks 9Router's LLM API to strangers

A GitHub advisory says 9Router treats any request carrying an X-9r-Real-Ip header of 127.0.0.1 as local, skipping the API-key check entirely.

The Agentic Times · 22 Sept 2026

119

A fortress wall with three identical breach points stands exposed as guards repeatedly fail to verify the true identity of approaching messengers before granting passage to forbidden inner chambers.

Vulnerabilities

CKAN MCP server SSRF guard bypassed for the third time

An advisory published on 22 September 2026 says the CKAN MCP server's URL check never resolves DNS, letting callers reach loopback and cloud metadata endpoints.

The Agentic Times · 22 Sept 2026

120

A mechanical hand reaches through a locked vault door, depositing papers into forbidden chambers throughout a sprawling archive while guards sleep.

Vulnerabilities

NotebookLM MCP server patched over arbitrary file write flaw

A path traversal bug in the vault batch tool let callers, or a prompt-injected model, write files anywhere the server process could reach.

The Agentic Times · 22 Sept 2026

121

Few AI-found bugs from Anthropic's Glasswing get exploited

Vulnerabilities

Few AI-found bugs from Anthropic's Glasswing get exploited

VulnCheck researcher says just one of 225 CVEs credited to Anthropic or its Project Glasswing partners has been attacked in the wild.

The Agentic Times · 21 Sept 2026

122

Researcher shows Meta Muse flaw can hijack dictation on macOS

Vulnerabilities

Researcher shows Meta Muse flaw can hijack dictation on macOS

A proof-of-concept published on 21 September 2026 shows local malware can point the AI assistant's dictation traffic at an attacker's server.

The Agentic Times · 21 Sept 2026

123

BragJack demo hijacks AI browser agents via one extension

Vulnerabilities

BragJack demo hijacks AI browser agents via one extension

A proof-of-concept attack from Forever Security targets AI assistants in Chrome, Edge, Opera Neon, Perplexity Comet and Claude in Chrome, earning two CVEs.

The Agentic Times · 19 Sept 2026

124

A locked vault's three doors swing open simultaneously—one revealing stolen keys, another exposing scattered documents, a third bypassing the guard entirely.

Vulnerabilities

Three flaws in Obot MCP platform let attackers steal API tokens

Advisories published on 18 September 2026 describe token theft, an unauthenticated registry leak and server-side request forgery, all fixed in Obot v0.23.0.

The Agentic Times · 18 Sept 2026

The morning agent security brief

One lead story, four links, the week's incident count. Weekday mornings. Free.