CERT-In's advisory on AI agents as attackers: what to fix in your own estate
CERT-In says agents under evaluation reached Indian firms' live systems through exposed credentials. What the advisory asks of teams that run agents.
5 min read
Blog tag
#Security
Threats, controls and secure design for AI tools and the systems they touch.
What a low prompt-injection success rate doesn't tell you about an agent
New research finds low attack success rates mostly mean agents ignored injections, and detector rankings don't transfer. How to read these evaluations.
5 min read
FastMCP's October advisories: which versions to run and what upgrading misses
FastMCP published 16 advisories in five days, including client-side code execution. The versions that fix them, and the steps an upgrade won't do.
Anthropic user profiles: keeping one customer's ban off your whole platform
Anthropic's new user profiles let platforms attribute API requests to end customers, so enforcement can hit one tenant. What to send, and what is not live yet.
Step-up authorisation in MCP: start agents with minimal scopes
A large MCP server now lets agents connect with minimal OAuth scopes and ask for more mid-task. How the specification's step-up flow works, and what to check.
5 min read
GitHub's package firewall preview: a checkpoint between coding agents and npm
GitHub's package firewall checks npm installs from CI runners and coding agents against age, malware and attestation rules. How it works, and its limits.
6 min read
MCP servers over HTTP: the SDK resource limits you now have to set
Recent MCP SDK advisories added size, batch, session and validation limits to HTTP servers. Some need a code change, not just an upgrade. What to check.
Reflection's Beam: what a 501B open-weight model's safety report tells agent teams
Beam arrived on 16 October under Apache 2.0 with a long safety report. Its agent results show which safeguards come with the weights and which don't.
Copilot managed settings can now fail closed: forced refresh and minimum versions
Copilot clients can now be made to fetch current policy at startup and refuse to run on a stale copy. What that buys, what it costs, how to roll it out.
An MCP login link that ran code: opencode's fix shipped without an advisory
A remote MCP server could run commands on Windows and WSL machines through opencode's OAuth flow. The fix is out, but no CVE or advisory will flag it.
5 min read
Open-weight model risk, model layer first: the Microsoft and NVIDIA guidance
Microsoft and NVIDIA published a joint risk taxonomy for open-weight models. What it covers, what it leaves for later, and how coding teams can use it.
Agent skills that point at bad domains: what a 5.5 million URL scan found
Manifold checked every URL in public agent skills, MCP servers and plugins; 121 components use a high-risk domain. What that means for reviewing skills.
Browse by tag
- #Security (54)
- #Agents (40)
- #AI coding tools (37)
- #LLM costs (23)
- #MCP (16)
- #Evaluation (14)
- #Observability (12)
- #Compliance (11)
- #Enterprise (9)
- #Platform engineering (9)
- #Data privacy (8)
- #DevOps (7)
- #Model routing (7)
- #Open-weight models (7)
- #Research (7)
- #Tutorials (6)
- #Code review (4)
- #Developer productivity (4)
- #Guardrails (4)
- #OpenTelemetry (4)
- #Budgets (3)
- #Networking (3)
- #Prompt engineering (3)
- #Self-hosting (3)
- #MCP gateway (2)
- #Release notes (1)
- #Skills (1)
- #Testing (1)