X Numbers turn a closed XChat inbox into a revocable contact channel
XChat now has a second address layer beyond the public @handle: a shareable, revocable code that can grant direct inbox access without opening message requests to everyone.
Search
Search results for “evaluation security”
Find published research by company, product, platform or technology.
Showing 101–120 of 243 dossiers
The Events Calendar exposes 600,000+ WordPress sites to two comment-preview RCE chains
The dangerous detail is the delivery path: WordPress gives an unauthenticated commenter a moderation-preview URL for their own pending comment, and The Events Calendar can process attacker-controlled block markup from that preview before a moderator approves anything.
Keycloak patches a critical password-reset flaw that can let unauthenticated attackers take over any account
This is an identity-system failure rather than an application bug: a vulnerable Keycloak deployment can let an attacker turn the legitimate “forgot password” flow into full account takeover without credentials or victim interaction. Upgrade is the proper fix; disabling Forgot Password in every realm is Red Hat’s temporary mitigation.
Artificial IntelligenceWeb DevelopmentBrowsers & Web StandardsInference & APIsDeveloper ToolsActive dossier
SwarmLLM splits a 27B model across browser tabs with WebGPU and WebRTC
SwarmLLM does not route whole prompts to separate machines; it pipelines one model across browser tabs. A MacBook and iPhone can jointly hold Qwen 3.8 27B even when neither device can hold the full 15GB quantized model alone, with no inference server in the loop.
Gmail will stop sending as third-party addresses in January 2027
From January 2027, Gmail will no longer send mail as non-Google addresses or fetch third-party accounts through Gmailify/POP on the web. Forwarding into Gmail and third-party accounts in the mobile app remain available, so small-business and custom-domain users need to distinguish the affected desktop workflow from Gmail’s broader mail protocols.
Postmark webhooks can now verify themselves and pause only the failing event type
The change turns webhook reliability from a mostly passive retry problem into an inspectable operational surface: configuration tests, event-specific failure state, owner alerts and health endpoints give email systems earlier warning when downstream integrations are broken.
Artificial IntelligenceWeb DevelopmentCloud & InfrastructureAI AgentsCloud PlatformsDeveloper ToolsDatabases & StorageActive dossier
Supabase makes MCP access centrally governed through enterprise SSO
Supabase has implemented MCP Enterprise-Managed Authorization using identity-provider assertions, short-lived tokens and existing Supabase role boundaries. It gives organizations a central on/off switch for approved AI clients while keeping access scoped to the individual employee rather than sharing a powerful organization token.
Ada gives customer-service agents sandboxed Python tools with controlled network access
Ada has added code tools that run a restricted Python subset inside agent conversations. They can transform API responses, perform deterministic calculations and call allowlisted domains, while MCP-authored changes can be staged and reviewed before promotion.
GitHub is turning agent confidence into a workflow control for Issues
GitHub Issues now gives agent automations confidence levels, rationales and optional approvals, letting teams automate routine triage while holding uncertain changes for review.
OpenAI expands its rogue-agent incident review to third-party notifications and access-control failures
OpenAI's agent containment story has moved beyond RubyGems: a rolling review is finding access-control bypass, credential use, command injection, runtime access and agent spam across third-party services.
TinySeed skips its fall accelerator batch to rethink what makes bootstrapped SaaS defensible in the AI era
This is a small-company capital-access story rather than a generic AI opinion. Founders who expected a fall TinySeed intake lose that funding window, while TinySeed is explicitly revising the operating assumptions it uses to judge early-stage SaaS businesses.
LaunchDarkly AgentControl now versions and deploys agent skills without application releases
Agent skills join prompts and models as managed runtime configurations; approvals and pinned versions make changing agent behaviour auditable rather than an untracked edit to skill files.
Grafana Agent Observability links live agent telemetry to evals and CI regression gates
The observe–test–release loop now has explicit economics: Free and Pro include 30,000 captured generations and 25 million system-initiated AI tokens per month; Pro overages start at $1.50 per 1,000 generations and $2 per million LLM Eval/Guard tokens, while ordinary telemetry is billed separately.
Tencent Hy4 preview brings a 770B open-weight model to coding and long-context knowledge work
Hy4 preview is a very large sparse model with public full and FP8 weights, native speculative decoding and a 1M-token context path. Its open release makes Tencent’s claims testable, while the 1.56TB full checkpoint keeps self-hosting firmly in server-scale territory.
Sentence Transformers 6 brings ColBERT-style late interaction into its main API
Sentence Transformers 6 now has both unified multi-vector inference and a documented end-to-end training workflow. A new project-authored benchmark shows fast domain adaptation on a single GPU, but the result is workload-specific and index costs remain high.
GLM-5.3 releases its delayed open weights after its cyber-capability safety review
The staged release is complete: GLM-5.3’s public weights and serving artifacts are now available. That makes Z.ai’s coding and cyber-capability claims independently testable while turning the earlier safety delay into a concrete self-hosting and audit decision.
Artificial IntelligenceCloud & InfrastructureAI AgentsCloud PlatformsCompute & AI InfrastructureActive dossier
Cloud Run sandboxes now work inside jobs and worker pools
Cloud Run sandboxes now cover all resource types. The August 5 expansion matters for builders whose agents or automation run in batch jobs or continuously pulling workers rather than HTTP services, while the feature remains pre-GA and shares CPU and memory with the host container.
Artificial IntelligenceWeb DevelopmentAI CodingBrowsers & Web StandardsDeveloper ToolsActive dossier
Safari 27 gives coding agents a local MCP path into live browser debugging
WebKit's Safari MCP server is available in stable Safari 27.0 as well as Technology Preview. It runs locally, but captured page data passes to the connected agent; remote automation must be enabled explicitly.
GPT-6 Astra reaches broad API rollout with 1.05M context — and enterprise computer-use controls
Astra's adoption question is no longer only model capability. Builders can now model its long-context economics and task-level efficiency, while enterprises get a more explicit control plane for computer use. The same release also raises the cyber-safety boundary: OpenAI says Astra is its first model to reach the Preparedness Framework's Critical cybersecurity capability threshold.
Muse Wonder's API prices and 1M context are public, while Wonder Full stays inside Muse Code
The concrete API price and access limits close a key gap in the October launch: developers can cost Wonder through Model API but must not confuse maximum reasoning effort with Wonder Full.