Evidence search
Search
Search titles, summaries, topics, providers, authors, and the full open-access corpus.
Results for “Architecture”
68 publicationsRun AI coding agents in GitHub Actions without exposing secrets
Treat the agent job as an untrusted-input processor. Follow the outside text from the trigger that admits it to the credentials it can reach, then split reading from writing so a successful injection reaches nothing worth stealing.
AI systems · GitHub / Anthropic / Google / Microsoft · By Cloud Security DeskTreat a cloud model retirement as a security change
Bedrock and Agent Platform fail closed on a model's retirement date, while Foundry can swap a different model in behind the same deployment name. Read the three calendars, compare the replacement's controls and decide per deployment.
AI systems · Amazon Web Services / Microsoft Azure / Google Cloud · By Cloud Security DeskInvestigate a SaaS integration breach through its OAuth tokens
A vendor that stores OAuth tokens for your SaaS tenants has been breached. Confirm from your own logs what the tokens did, revoke them on every platform, and search the exposed records for secrets.
Detection & response · Salesforce / Google Workspace / Microsoft Entra ID · By Cloud Security DeskFind the AWS global control planes in your recovery path
Thirteen AWS control planes run in us-east-1 or us-west-2, and ordinary calls from other Regions still reach them. Audit runbooks, SDK settings and the last exercise, then swap each dependency for a data-plane operation.
Resilience · Amazon Web Services · By Cloud Security DeskLock down self-hosted model servers before exposing them
Built-in credentials in vLLM, SGLang and Triton cover less than their names suggest, Ollama's local API has none, and vLLM and SGLang leave their internal planes unauthenticated. Here is what stays open and what must sit in front.
AI systems · vLLM / SGLang / Ollama / NVIDIA Triton Inference Server / Kubernetes / nginx · By Cloud Security DeskPlan for an outage at your CDN or edge provider
When the CDN, WAF and edge provider fails, traffic can move to a regional WAF, a second CDN or the origin, and each keeps less protection. Decide early whether a second path is worth securing.
Resilience · Microsoft Azure / Cloudflare / Amazon Web Services / Google Cloud · By Cloud Security DeskUse DNS egress filtering without mistaking it for a firewall
Resolver-layer filters in AWS, Azure and Google Cloud decide which names a workload can resolve, not where it connects. What each provider says its DNS control misses, and the connection control to pair with it.
Workload security · Amazon Web Services / Microsoft Azure / Google Cloud · By Cloud Security DeskKeep semantic response caches inside authorization scope
A semantic cache returns whatever its nearest stored prompt earned. Put tenant, user or entitlement scope, template and model into an exact-match partition, close the write path, and treat thresholds as a correctness setting.
AI systems · Microsoft Azure / Google Cloud / Redis / GPTCache / Amazon Web Services · By Cloud Security DeskCompare just-in-time privileged access in AWS, Azure and Google Cloud
Entra PIM and Google Cloud PAM make roles temporary natively, while AWS relies on TEAM or partner tools. All three end elevation by removing an assignment, and sessions, caches and minted tokens can keep running.
Identity & access · Amazon Web Services / Microsoft Azure / Microsoft Entra / Google Cloud · By Cloud Security DeskPlan Azure disaster recovery for a region without a pair
Thirteen Azure regions have no pair and eight more pair with a restricted-access region. This guide shows which built-in geo-redundancy disappears, what replaces it service by service, and how to choose the recovery region.
Resilience · Microsoft Azure · By Cloud Security DeskDesign AI agents that contain indirect prompt injection
Filters lower the odds that an agent obeys injected text. Architecture decides what an obeying agent can reach. Compare six published patterns and CaMeL by what each removes, what it costs and what it still misses.
AI systems · Google DeepMind / AgentDojo / NIST / OWASP / Microsoft · By Cloud Security DeskContain failures with cell-based architecture and shuffle sharding
Cells limit a failure to the tenants behind one partition only when routing, data and change delivery are partitioned too. Here is how to choose the key, size cells and check the shuffle-sharding arithmetic.
Resilience · AWS / Microsoft Azure / Slack · By Cloud Security DeskKnow where cloud AI services process your prompts
Bedrock inference profiles, Foundry deployment types and Agent Platform endpoints each decide where a prompt is processed, separately from where data rests. Compare what each provider commits to and how to enforce it.
AI systems · Amazon Web Services / Microsoft Azure / Google Cloud · By Cloud Security DeskCompare managed prompt attack filters in Bedrock, Azure and Vertex AI
Bedrock Guardrails, Azure Prompt Shields and Google Cloud Model Armor all score text for injected instructions. They differ in which content reaches the classifier, how you mark it, and what each one skips.
AI systems · Amazon Web Services / Microsoft Azure / Google Cloud · By Cloud Security DeskBuild a cryptographic inventory for the post-quantum transition
Record where public-key cryptography protects long-lived data and which library, service or vendor must change it. Then rank entries by function and data lifetime, not by how many RSA keys a scanner found.
Workload security · NIST / CISA / Office of Management and Budget / Amazon Web Services / Google Cloud / Microsoft Azure · By Cloud Security DeskRun code written by AI agents inside a disposable sandbox
Treat agent-generated code as untrusted input. Pick a microVM or user-space kernel boundary, keep egress and credentials off by default, and destroy the session when the user or conversation ends.
AI systems · Amazon Web Services / Microsoft Azure / Google Cloud / Firecracker / gVisor / OWASP · By Cloud Security DeskScan uploaded files for malware before your application trusts them
Hold every upload where nothing can read it, promote only files with a clean verdict, and decide in advance what happens to the files GuardDuty, Defender or a ClamAV deployment cannot scan.
Workload security · Amazon Web Services / Microsoft Azure / Google Cloud / OWASP · By Cloud Security DeskGive AI agents their own identity instead of borrowed user tokens
An agent that replays a user's token is invisible in logs and cannot be revoked on its own. Give it a principal, delegate narrowly through token exchange, and use what Entra, AgentCore and Google now provide.
Identity & access · Microsoft Entra / Amazon Web Services / Google Cloud / IETF / Model Context Protocol · By Cloud Security DeskPrepare a cloud exit strategy that satisfies DORA
DORA requires a tested exit plan for every cloud service that supports a critical or important function. This guide turns that duty into dependency maps, measured data exports, staged tests and register entries that agree.
Resilience · European Commission / European Supervisory Authorities / European Central Bank / Amazon Web Services / Google Cloud / Microsoft Azure · By Cloud Security DeskKeep an S3 bucket private with Block Public Access
Enable S3 Block Public Access while preserving approved readers. Review the four settings, private CloudFront origins, and practical access tests.
Workload security · AWS · By Cloud Security DeskRemove public SSH access from an EC2 security group
Remove internet-wide EC2 SSH rules after testing Session Manager or restricted SSH. Check all groups, IPv6, fresh sessions, and recovery access.
Workload security · AWS · By Cloud Security DeskRequire authentication between Cloud Run services
Configure Cloud Run service-to-service authentication with a dedicated caller identity, the right ID token audience, and useful negative tests.
Workload security · Google Cloud · By Cloud Security DeskCheck and stop anonymous access to Azure blobs
Understand account and container settings and prove both anonymous denial and intended app access.
Workload security · Azure · By Cloud Security DeskHandle Azure OpenAI rate limits without retry storms
Separate deployment quota, short request bursts and bounded client retry behavior.
AI systems · Azure · By Cloud Security Desk