Salt Introduces Prompt Security with new AI-DR Solution
Language

Security Intelligence

Hub API and Agentic
Threat Monitoring

0
Stories tracked
1
Days covered
June 20, 2026
Oct 5, 2026
Latest brief

Threat archive

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
October 5, 2026

Apple to Tighten macOS Full Disk Access as AI Agents Gain Broad System Reach; Researchers Show Token Theft in Meta Muse and ChatGPT Mac Apps

Apple said it will require explicit user action to grant Full Disk Access, warning that risks grow substantially as AI agents become more capable and autonomous. The move follows reports of Meta's Muse agent reading a journalist's iMessages after FDA was granted, plus Patrick Wardle's PoC exposing a Muse dictation-endpoint flaw enabling auth-token capture and prompt injection, and CVE-2026-100754 in OpenAI's ChatGPT for Mac.

Agents inherit privileged access to local data and backend APIs, making their tokens and connectors high-value targets. Enterprises need visibility into agent permissions and the APIs those agents call on users' behalf.

Agentic/MCP
Read full story
October 4, 2026

China-Aligned TA419 Targets U.S. AI Policy Experts With Frameless Browser-in-the-Browser AitM Phishing

Proofpoint attributed credential phishing campaigns against AI experts at U.S. think tanks, universities, and law firms to China-nexus actor TA419. The group impersonated economists, a former White House OSTP leader, and an Anthropic employee, using benign first-contact lures before sending shortened URLs leading to a OneDrive-themed Frameless BitB page. A custom telemetry module tracks the Microsoft sign-in flow and proxies credentials to real Microsoft infrastructure, silently capturing session cookies.

AitM proxies capture live session tokens, turning phishing into authenticated API access that bypasses MFA. Detecting anomalous token reuse against cloud APIs matters as much as phishing-resistant passkeys.

Breach
Read full story
October 3, 2026

Fortra Patches Critical BoKS Flaws: Predictable AD Service Account Passwords and Root Command Injection Reachable via REST/SOAP API

Fortra fixed eight Core Privileged Access Manager (BoKS) vulnerabilities, including CVE-2026-79901 (CVSS 9.9), where Active Directory service account passwords are generated from a pseudo-random sequence seeded with the Unix timestamp, enabling offline candidate verification with Kerberos ticket material. CVE-2026-79898 (CVSS 9.1) is a crlserver command injection exploitable through BCC and the WSI REST or SOAP API that runs as root on the BoKS Master, alongside an autoregistration stack buffer overflow (CVE-2026-12627).

A root command injection reachable over the network through a privileged access manager's REST and SOAP APIs shows why PAM control-plane APIs need the same scrutiny as public ones.

API Security
Read full story
October 2, 2026

Fortinet FortiMail Zero-Day CVE-2026-104286 Exploited for Unauthenticated Arbitrary File Writes; Added to CISA KEV

CISA added CVE-2026-104286 (CVSS 9.8) to its KEV catalog after confirmed in-the-wild exploitation. The path traversal plus NULL byte handling flaw lets unauthenticated attackers write arbitrary files on FortiMail via crafted HTTP/HTTPS requests. Fortinet advised disabling the IBE encryption feature and restricting management interface access until patches ship, and published attacker IPs and implanted file indicators such as /data/etc/ld.so.preload and a modified httpd.conf.

Unauthenticated HTTP request handling flaws in edge appliances are API-layer attack surface reachable from the internet. Inventorying and restricting exposed management APIs limits exploitation windows before patches exist.

Breach
Read full story
October 2, 2026

Dell Container Storage Modules Flaws Include CVSS 10.0 Missing Authentication on gRPC and Authorization Proxy APIs

Dell patched multiple critical CSM vulnerabilities, including CVE-2026-63688 and CVE-2026-63692 (both CVSS 10.0), where missing authentication on the csm-authorization-storage gRPC server and the authorization proxy/tenant service let unauthenticated remote attackers retrieve storage backend admin credentials or gain administrative privileges. Hard-coded credentials and signing keys (CVE-2026-54472, CVE-2026-61421) allow forging valid admin JWTs. All versions before 1.17.0 are affected; fixed in 1.18.0 with no workarounds.

Unauthenticated gRPC endpoints and hard-coded JWT signing secrets are classic API authentication failures in infrastructure control planes. API discovery and posture checks surface these exposed, unauthenticated services before attackers do.

API Security
Read full story
September 30, 2026

Cisco Catalyst SD-WAN Manager Zero-Day (CVE-2026-76504) Exploited to Hit Management API as Admin

Cisco disclosed active exploitation of CVE-2026-76504 (CVSS 9.8), a URI-encoding flaw in Catalyst SD-WAN Manager that lets an unauthenticated remote attacker bypass an authentication rule and reach the Manager's API with admin (netadmin) privileges. Fixed releases are available with no workaround; CISA added the flaw to its KEV catalog with an October 3, 2026 federal deadline. Cisco published log-based IoCs tied to encoded /j_security_check requests.

An encoding trick that bypasses an auth rule in front of one API endpoint is a textbook API authorization failure. Runtime API behavioral monitoring catches admin-level API abuse that signature-based controls miss.

API Security
Read full story
September 30, 2026

Zimbra Flaw CVE-2026-73570 Exploited to Deploy Web Shells and Harvest Authentication Secrets Including PreAuth and 2FA Keys

Microsoft reported active exploitation of CVE-2026-73570 (CVSS 8.9), an unauthenticated OS command injection in Zimbra Collaboration triggerable via a crafted SMTP request when zimbra-snmp and SNMP notifications are present. Attackers deployed JSP web shells and reverse shells, escalated privileges via /etc/pam.d/sudo, and used zmlocalconfig plus LDAP queries to steal zimbraPreAuthKey, zimbraAuthTokenKey, and zimbraTwoFactorAuthSecret, then exfiltrated mailbox archives using AzCopy to attacker cloud storage.

Harvesting pre-auth and token signing keys lets attackers mint valid sessions against mail APIs indefinitely, so patching alone is insufficient. Secret rotation plus monitoring for anomalous authenticated API access is required.

Breach
Read full story
September 29, 2026

Official MCP Python SDK Flaw Lets Malicious MCP Servers Steal OAuth Client Secrets and Authorization Codes

Affected versions of the official MCP Python SDK failed to validate which authorization server an MCP server pointed clients to, letting a malicious server collect the client secret, authorization code, and PKCE proof key. Cycode demonstrated exchanging those credentials for a valid access token carrying the app's granted permissions. Fixes landed in versions 1.30.0 and 2.2.0, though ClientCredentials and PrivateKeyJWT providers also require setting issuer=.

Agent toolchains hold real OAuth credentials for production APIs, so an untrusted MCP server becomes an API credential-theft channel. Discovering MCP clients and monitoring token issuance and API usage is essential.

Agentic/MCP
Read full story