Skip to content

Topic

AI Application Security

Practice of securing AI and LLM-based applications, covering risks like prompt injection, data leakage, and unsafe handling of model inputs and outputs.

Current stories

securityOne report1 publisher

Langflow's CVSS 9.9 MCP flaw has drawn 1,423 honeypot exploit attempts from 27 IPs

Attackers made 1,423 attempts against Previdian honeypots since September 19 to exploit Langflow CVE-2026-105740 (CVSS 9.9, versions before 1.9.0). In one sequence they got the required login with a common username and password, Previdian said.

Reality

Evidence65
Adoption
Not scored
Hype gap−5
Incentives40
Confidence65
productOne report1 publisher

HackerOne's data shows open vulnerability backlogs outrunning a halved time-to-fix

HackerOne says companies cut average vulnerability fix time from 135 days to 62, yet validated findings left unresolved rose 131% over two years. That leaves time-to-fix a poor guide to exposure, and paying the debt down means taking capacity away from feature work.

Publishers:devops.com

Reality

Evidence45
Adoption50
Hype gap+25
Incentives70
Confidence45
buildOne report1 publisher

Flowise deleted its CSV Agent after the regex guarding model-written code failed repeatedly

Flowise fixed a CVSS 9.8 remote-code bug in its CSV Agent with a stricter regex, then saw that filter fail through five more CVEs. It deleted the feature in June, and the patch history is a case against trusting a string filter to vet model-written code.

Publishers:dev.to

Reality

Evidence64
Adoption
Not scored
Hype gap−5
Incentives25
Confidence60
securityOne report1 publisher

DARPA picks AIxCC winner Xint to research AI audits of military messaging apps

DARPA selected Xint, one of three winners of its $29.5 million AIxCC contest, to research AI-driven security analysis of Department of War messaging apps. The work aims its automated find, triage and patch pipeline at messaging code and binaries the military actually runs.

Publishers:securityweek.com

Reality

Evidence35
Adoption25
Hype gap+40
Incentives70
Confidence40
securityConfirmed4 publishers

Detections on VulnCheck's canaries climb from 50 to 360 amid Langflow, Rails exploitation

The probes read Langflow's secret key file and grep the process environment for OpenAI and AWS credentials, which puts an AI orchestration tool on the same scanning clock as the Rails file-read bug of the same week.

Perspective Coverage

4 publishers
Builder
Builder 30%
Operator
Operator 60%
Investor
Investor 10%

Reality

Evidence62
Adoption40
Hype gap+15
Incentives65
Confidence60
buildOne report1 publisher

A RAG design re-reads the user's department from Postgres before every vector search

A dev.to write-up keeps access rights in Postgres and vectors in Qdrant, with a department_id stamped into every chunk payload so the filter runs before the model sees any text. The token supplies the identity, and the department comes from Postgres.

Publishers:dev.to

Reality

Evidence55
Adoption
Not scored
Hype gap+20
Incentives30
Confidence58
buildOne report1 publisher

Six to nine vendors, five obligations each: the first AI security exercise is arithmetic

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.

Publishers:dev.to

Reality

Evidence54
Adoption
Not scored
Hype gap+22
Incentives78
Confidence46
buildOne report1 publisher

A docs bot that refuses to answer is working: the case for an evidence gate over a bigger window

An architecture decision record for a gaming moderation assistant puts a policy stage between retrieval and generation, and counts abstention with a reason code as a successful outcome.

Publishers:dev.to

Reality

Evidence30
Adoption
Not scored
Hype gap+18
Incentives25
Confidence34