Standard
OWASP's ranked list of top security risks in large language model applications, covering issues like prompt injection, excessive agency, and supply chain flaws.
Apple announced on October 2 that macOS Full Disk Access can be granted only through very explicit user action, citing the added risk from AI agents. Agent builders can scope to user-selected folders now.
Reality
PwC surveyed 3,934 leaders and found 52% rank attacks on AI systems as their biggest cyber preparedness gap. Only 17% place AI risk with the CISO, so in most of the sample the security function does not own the threat it feels least ready for.
Perspective Coverage
3 publishersReality
Dev.to author dharani2d argues agent security depends on who picks the next tool call, citing Excessive Agency's rise from sixth to third at OWASP. The proposed control plane keeps identity, authorization, argument checks and approvals in deterministic code outside the model.
Reality
AegisGate's founder replayed 24 attack prompts gathered from advisories and disclosed incidents against his own gateway, which scores 99.8 out of 100 on evasion resistance. It blocked 13, and seven new regexes followed.
Reality
Only 36% of IT staff in an Automox survey feel highly confident in their endpoint compliance visibility, as AI agents start acting on managed devices. Agents inherit their launcher's rights, so teams now need a record of what each may do and what it did.
Reality
OWASP's August 3 edition added no categories and removed none, yet eight of the ten entries changed rank, with Unbounded Consumption up four places and Improper Output Handling down five. Prompt Injection still holds first.
Reality
Forcepoint's unguarded research agent ran 500 tool calls for a simulated $10, a total set by the test's own ceiling, while the same scenario with a call budget, recursion cap and circuit breaker ended after one call and $0.02.
Reality
Microsoft and Omdia put the average security operations centre at close to 3,000 alerts a day with 42% never investigated, while a 2026 benchmark scored the best frontier model at 3.8% on finding malicious events in raw Windows logs.
Reality
A dev.to analysis of the Model Context Protocol argues the tool layer is where an attacker gets to write instructions, and that the agent then carries them out holding credentials the content author never had.
Reality
Forcepoint X-Labs hid instructions in zero-pixel white HTML and watched a summarizer repeat them as fact. The measurement is new; two named practitioners say the design flaw it measures has been documented since 2023.
Reality
An SC World commentary argues the AI security debate is stuck on model behavior while agent connectors get wired with API keys that never rotate and model loading still executes unsigned code.
Reality
OWASP's v1.0 standard says the payload is prose and the permissions belong to the host agent, which is another way of saying the SAST and SCA licences already on the books scan straight past the thing teams are installing this quarter.
Reality
A vendor-authored count says a typical production AI app routes through 6 to 9 services, each adding a key, an egress path, an in-runtime SDK, a prompt log and a subprocessor. The count holds up.
Reality