The UnboundCompute blog
Writing on web security
-
The Single Packet Attack: Making Web Race Conditions Reliable
The single packet attack removes network jitter so many requests hit the same check to action window at once, making a flaky…
-
MCP Command Injection: When a Tool Argument Becomes a Remote Shell
MCP command injection turns a tool argument into an executed shell command. Learn how prompt injection triggers it remotely and how to…
-
The SAML Authentication Bypass Behind SAP NetWeaver
A SAML authentication bypass lets a valid signature cover one element while the server reads identity from another. See how SAP’s flaw…
-
OIDC Authentication Bypass: Trusting a Token You Never Verified
An OIDC authentication bypass lets a forged JWT log in as an admin when an app reads token claims but never verifies…
-
Limit Overrun Race Conditions Explained
A limit overrun race condition lets an attacker fire parallel requests to beat a check then act window and redeem one coupon…
-
JWT Algorithm Confusion Attacks Explained
JWT algorithm confusion lets an attacker pick how a token is verified. See how the none bug and RS256 to HS256 key…
-
Broken Function Level Authorization (BFLA) Explained
Broken function level authorization lets a normal user call an admin function directly. See how BFLA differs from BOLA, with concrete API…
-
Broken Object Property Level Authorization (BOPLA) Explained
Broken object property level authorization (BOPLA, API3:2023) lets users read hidden fields or write privileged ones. See how it works and how…
-
Agent Delegation Limits as a Defense
Agent delegation limits cap how deep, how wide, and how expensively an agent can spawn sub agents, stopping runaway loops and swarms…
-
Agent in the Middle Attacks Explained
An agent in the middle attack lets an attacker edit messages between two real agents in transit. See how it works and…