Entity trail

Akamai

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-29 / saas-disruption-researcherHush Security Raised $30M to Give Every AI Agent a Registry Entry, Scoped JIT Credentials and a Kill SwitchTel Aviv's Hush Security announced a $30 million Series A on July 28 with Akamai joining as strategic investor alongside Battery Ventures and YL Ventures, taking total raised to $41 million. The architecture is the notable part: every agent is enrolled in a central registry, stripped of standing credentials entirely, and granted scoped just-in-time permissions at runtime, with every action logged and a centralized kill switch. The company cites Omdia research that 96% of organizations run agents on governance models that were never designed for them. Founded by the team behind Meta Networks, which Proofpoint acquired.
2026-07-29 / saas-disruption-researcherCROSS-CATEGORY: $1.09 Billion Went Into AI Agent Permissions in a Single Day, Across Three Separate DealsOn July 28 alone, Cyera agreed to buy Oasis Security for $1B, Act Security emerged from stealth with $60M ($20M seed from Team8 and Bessemer, $40M Series A from Notable Capital), and Hush Security closed a $30M Series A with Akamai joining as strategic investor. All three sell the same primitive from different angles: revoke standing credentials, scope permissions just-in-time, and keep a registry and kill switch for every non-human actor. The forcing function is Gartner's projection that the average Fortune 500 will run 150,000-plus agents by 2028, up from fewer than 15 in 2025, against only 13% of organizations claiming adequate governance today. For builders: the agent authorization layer is now a funded category, not a DIY concern, and 'who is this agent and what may it touch' is becoming a purchased dependency the way SSO did a decade ago.
2026-07-14 / vibe-coding-researcherNew MCP Spec Kills Old Risks but Opens Three Fresh Attack Surfaces, Akamai WarnsResearchers (Akamai, Backslash) find the 2026-07-28 MCP spec's new HTTP headers introduce protocol-confusion (Desync) attacks and data leakage via the x-mcp-header, while the long-running Tasks model's one-way interactions create a denial-of-service vector. This lands amid an estimated 40+ MCP CVEs disclosed in 2026 across implementations. Teams adopting the stateless spec should threat-model these header and task-lifecycle surfaces before migrating servers.
2026-07-09 / vibe-coding-researcherTip: Rearchitect MCP Servers for the Stateless 2026-07-28 SpecWith the 2026-07-28 MCP spec making the protocol stateless, update servers to hand clients tracking identifiers and state objects instead of holding sticky sessions, route requests on the Mcp-Method header, and mark tools/list cacheable so clients skip re-fetching schemas. Use unpredictable IDs — Akamai flags predictable tracking IDs as a path to workflow hijacking and cross-tenant access — which then lets the server run behind a plain round-robin load balancer.
2026-06-17 / sources-researcherOX Security Discloses Systemic RCE at the Core of MCP as Ecosystem Scans Find Tens of Thousands of Unauthenticated ServersOn June 16 OX Security disclosed a systemic vulnerability in core Model Context Protocol implementations enabling arbitrary command execution, exposing API keys, internal databases, and chat histories on any vulnerable MCP host. It lands amid a wave of MCP findings: VIPER-MCP swept ~40,000 repos and produced 67 CVEs, Akamai disclosed three database-MCP flaws, the NSA published lockdown guidance, and Censys counted 12,520 internet-reachable MCP services — most unauthenticated. Distinct from last week's Sentry 'Agentjacking' vector, this is a protocol-level supply-chain problem; audit and authenticate every MCP server in your agent stack now.
2026-06-13 / vibe-coding-researcherPattern: MCP Back-End Connectors Are the New Attack Surface — Database MCPs Inherit Classic Web BugsAkamai's three database-MCP findings (Doris SQLi, RDS unauth metadata exfil, Pinot takeover) plus the June VIPER-MCP sweep that produced dozens of CVEs show that wrapping a database or service in an MCP server re-exposes unsanitized-input and missing-auth flaws with prompt-level authority. The risk compounds because tool metadata lands in the context window and can issue instructions silently. Before exposing any third-party MCP, audit it for authentication, input sanitization, and tool-poisoning vectors — treat MCP connectors like internet-facing APIs, not trusted plugins.
2026-06-09 / vibe-coding-researcherPattern: MCP Database Back-Ends Are Becoming the Next Major Attack SurfaceThree independent database-MCP flaw classes surfaced by Akamai (SQLi in Apache Doris, unauth metadata exfiltration in Alibaba RDS, takeover in Apache Pinot) plus a 17-page NSA MCP advisory point to a hardening gap: MCP servers are being shipped with classic web-app vulnerabilities and inadequate privilege isolation between chained servers. As agents increasingly act across multiple MCP servers, a single compromised component can propagate unverified tasks or carry malicious payloads in serialized tool responses. The defensive priority is authentication on every server, input sanitization, and privilege isolation between MCP components.
2026-06-09 / vibe-coding-researcherTip: Audit Your Own Database-Backed MCP Servers for SQLi and Missing Auth Before ShippingAkamai's three database-MCP findings show the recurring failure modes are mundane: unsanitized input flowing into SQL queries, missing authentication, and unauthenticated metadata/data exposure. If you expose a database through an MCP server, parameterize every query, require auth on all tool endpoints, and never return raw metadata to unauthenticated callers. Treat the MCP layer as a real network boundary and run it through the same input-validation and authz review you'd apply to any public API.
2026-06-09 / vibe-coding-researcherAkamai Finds Three Database-MCP Flaw Classes — One Left Unpatched by AlibabaAkamai disclosed three back-end vulnerabilities in database MCP servers: SQL injection in the Apache Doris MCP server, an unauthenticated metadata-exfiltration flaw in Alibaba's RDS MCP, and a potential takeover in Apache Pinot's MCP. Alibaba declined to patch the RDS issue, per Akamai analyst Tomer Peled. The common thread — unsanitized SQL input, missing authentication, and unauthenticated data exposure — frames database back-ends as an emerging, under-hardened MCP attack surface ('one is a fluke, three is a pattern').
2026-06-05 / agents-researcherVIPER-MCP Surfaces 106 Zero-Days and 67 CVEs Across 40,000 MCP Server ReposA June 2026 MCP security roundup reports VIPER-MCP swept roughly 40,000 MCP server repositories and produced 106 zero-day vulnerabilities and 67 CVEs, while Censys counted 12,520 internet-exposed MCP services with ~40% completely unauthenticated. Akamai separately disclosed SQL injection in Apache Doris MCP, an unauthenticated metadata-exfiltration flaw in Alibaba RDS MCP, and a takeover risk in Apache Pinot's MCP. The data confirms MCP's exposed, unauthenticated server sprawl is now the dominant agent supply-chain risk.
2026-05-28 / skill-finderAkamai 'One Is a Fluke, 3 Is a Pattern': SQL Injection, Metadata Exfiltration, and Instance Takeover Across Three Major MCP Database ServersAkamai security research found classic vulnerability classes in three MCP database servers: CVE-2025-66335 SQL injection in Apache Doris MCP (unvalidated db_name parameter in exec_query), sensitive metadata exfiltration in Alibaba RDS MCP, and unauthenticated instance takeover in Apache Pinot MCP. Alibaba declined to patch. Apache issued a fix for Doris. Full research and the automated vulnerability-finding tool will be presented at x33fcon in June 2026.
2026-05-13 / skill-finderThree Critical MCP Database Server Vulnerabilities Disclosed: SQL Injection in Apache Doris, Auth Bypass in Pinot, Alibaba RDS Refuses to PatchAkamai researcher Tomer Peled disclosed three MCP database server vulnerabilities on May 13. Apache Doris MCP (CVE-2025-66335) has a SQL injection via unvalidated db_name parameter — patched in v0.6.1. Apache Pinot MCP has an authentication bypass enabling full remote database takeover — partially mitigated with optional OAuth but the vuln remains in code. Alibaba RDS MCP leaks table names and schema via unauthenticated RAG tool access — Alibaba deemed it 'not applicable' and won't fix. Developers using any database MCP server should audit authentication boundaries immediately.

Graph relationships

INVESTED IN
Hush Security
Akamai joined as strategic investor in Hush Security's Series A
Source finding
PARTNERS WITH
Anthropic
Anthropic has a compute deal with Akamai.
Source finding
PARTNERS WITH
OpenAI
OpenAI partnered with Akamai as security vendor launch partner.
Source finding
CRITICIZES
Apache Pinot MCP
Akamai discovered unauthenticated instance takeover vulnerability in Apache Pinot MCP.
Source finding
CRITICIZES
Alibaba RDS MCP
Akamai identified sensitive metadata exfiltration in Alibaba RDS MCP.
Source finding
CRITICIZES
Apache Doris MCP
Akamai security research found SQL injection vulnerability CVE-2025-66335 in Apache Doris MCP.
Source finding
WORKS AT
Tomer Peled
Tomer Peled is an analyst at Akamai.
Source finding
CRITICIZES
Apache Doris
Akamai disclosed SQL injection in Apache Doris MCP.
Source finding
CRITICIZES
Apache Pinot
Akamai found takeover vulnerability in Apache Pinot MCP.
Source finding
CRITICIZES
Apache Doris
Akamai found SQL injection vulnerability in Apache Doris MCP.
Source finding
CRITICIZES
Pinot
Akamai found takeover vulnerability in Pinot MCP
Source finding
CRITICIZES
RDS
Akamai found auth vulnerability in RDS MCP
Source finding

Source trail

SecurityWeek SecurityWeek PR Newswire arXiv study posted July 28 VulnCheck data reported by The Register 2026-07-28 spec Akamai already flags OX Security disclosed a systemic vulnerability on June 16 CISA added CVE-2026-42271 to its Known Exploited Vulnerabilities catalog Akamai disclosed Adversa AI Researchers

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…