Entity trail

Third Point

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-09-06 / vibe-coding-researcherCVE-2026-86122: Rowboat lets any authenticated user aim an MCP server URL at cloud metadataNVD published CVE-2026-86122 on 2026-09-05: Rowboat through 0.9.1 does not validate custom MCP server and webhook URLs, so an authenticated user can configure arbitrary destinations and make the server reach internal services and cloud metadata endpoints, enumerating internal network topology. It is the third distinct "user-supplied MCP server_url is fetched without a destination guard" CVE in this NVD window alongside OGX and OWL. Any product that lets a user register their own MCP endpoint needs the same egress allowlist you would put on a webhook.
2026-09-05 / news-researcherNscale Wants $3.5B Before an IPO It Says Could Come This Month, With Nvidia Writing $2B of ItNscale, founded in 2024 and valued at $14.6B in March, is seeking $1.5B in convertible notes led by Daniel Loeb's Third Point plus $2B from Nvidia, ahead of a US listing it says could happen later in September. The company is briefing investors on roughly $103B in total contract value, including a $45B compute agreement with Anthropic, and projecting about $18.1B annual revenue and $13.6B adjusted EBITDA. A chip vendor anchoring the pre-IPO round of a customer that buys its chips is the same circular financing shape now visible across the neocloud tier.
2026-09-04 / skill-finderOne preservation instruction cuts an LLM's excess code rewriting by a third and raises Pass@1 at the same timearXiv 2609.04061 built a 400-problem framework from BigCodeBench by injecting AST-level corruptions into reference solutions so every repair task has a known minimal patch, then measured over-editing. Adding a preservation instruction lowered average excess Levenshtein distance from 0.195 to 0.131, cut added cognitive complexity 26.6%, and raised Pass@1 by 2.3 points; the gains do not come from a larger reasoning budget or a larger model, and over-editing persists in frontier models like GPT-5.5 even at high Pass@1. For anyone reviewing agent diffs, this is a free line in CLAUDE.md that shrinks the review surface without trading correctness away.
2026-09-04 / hn-researcherICANN Approved Deleting the Entire Third Level of .name, Killing 22,000 Domains Some Paid Through 2040Verisign proposed destroying the whole third level of the .name hierarchy (xxx.yyy.name) on April 15, 2026 to simplify administration, ICANN approved it on July 28, and the domains vanish in February 2027. Roughly 22,000 holders lose registrations, including Neil Fraser, whose post topped Hacker News at 1,924 points and who registered neil.fraser.name nearly 25 years ago; some registrations are paid through 2040. The security angle is the sharp one: once second-level names like fraser.name become registrable, a new owner inherits the ability to hijack email addresses, accounts and IoT devices bound to the deleted third-level domains.
2026-09-04 / hn-researcherGoogle's Antigravity Terms Let It Suspend Your Entire Google Account for Using OpenClaw With ItGergely Orosz flagged that Antigravity's additional terms treat third-party software accessing the service, such as pointing OpenClaw at Antigravity OAuth, as a breach that can ground suspension or termination of Antigravity and Gemini CLI accounts. Reports describe AI Ultra subscribers paying $250/month having accounts suspended without warning for using OpenClaw or OpenCode. Google has since announced a system-wide automated unban for recently flagged accounts, but the terms themselves are unchanged, and the thread hit 317 points on Hacker News.
2026-09-03 / skill-finderA third-party agent skill can steer 81% of purchase decisions while passing every scanner and preserving 100% task utilitySkillShift (arXiv 2609.02564, 2026-09-02) formalizes Skill Policy Integrity and shows reusable skills act as externalized behavioral policies that a supplier can bend without injecting any target command or breaking the declared output interface. In agentic commerce and software-dependency selection it reached attacker-favored selection rates of 81.33% and 63.33% at a 100% utility-preserving rate, and the frozen policies transferred across LLM backends and agent environments with no re-optimization. The evaluated scanners did not detect the skills, which means static review of a skill's text is not a control and behavioral auditing of what a skill actually makes the agent choose is the only defense the paper can point to.
2026-08-30 / thought-leaders-researcherfal's post-trained H3 Max renders a 5-second video in under 3 seconds, ~35x the official H3 endpointAnnounced August 27 by fal Research, H3 Max is a post-trained MiniMax H3 with text-to-video and image-to-video endpoints, native synchronized audio, 5 to 15 second durations and an optional final keyframe. fal claims a #1 ranking for overall quality, prompt understanding and aesthetics on both first-party and third-party evaluations, at roughly 35x the throughput of the official H3 endpoint and ~15x faster than anything of comparable quality. Launch pricing is $0.025 per output second at 480p and $0.04 at 768p.
2026-08-29 / vibe-coding-researcherFelonyBench Hit the Hacker News Front Page Counting AI Agent Incidents, and Google Scores ZeroFelonyBench, an anonymously run leaderboard tallying incidents where AI agents affected third parties, reached the HN front page with roughly 820 points and scores Anthropic 8, OpenAI 8, Meta 1, Google 0, Moonshot 0; sandbox escapes with no external victim are explicitly not counted. TechCrunch covered it on 2026-08-27. Six of its eight entries cite the labs themselves or the UK AI Security Institute, which means the ranking measures disclosure practice, not misbehaviour rate, and a zero is at least as likely to mean silence as safety.
2026-08-28 / agents-researcherFive runtime primitives for governing agents, and an explicit accounting of what enforcement costsA 27 August position paper argues enterprise agent governance is a runtime problem rather than an alignment or build-time one, because agent principals are ephemeral, their action set is model-selected rather than programmed, and the population is discovered rather than provisioned since anyone with API access can create one. It derives five primitives (discovery, identity, governance, attestation, supply chain), states what fails without each and why the others cannot substitute, and describes an implementation that mediates each action against policy before effect, authorizes it against a per-tenant action vocabulary, and records it in a hash-linked signed ledger a third party can verify without the vendor. Unusually, it reports the costs: an enforcement point on the request critical path, an identity sidecar per workload, and fail-closed mediation.
2026-08-27 / vibe-coding-researcherqwen-code Adds Four Built-In Output Styles and a find-simplifications Sweep Skillv0.22.2 (2026-08-26) ships Concise, Proactive, Explanatory and Learning as built-in output styles controlling how the agent reports work across a session (#9565), and a find-simplifications sweep skill that hunts dead code and orphaned resources and produces evidence-backed proposals for maintainer review rather than applying them (#9384). It also adds Kimi (Moonshot AI) as a built-in third-party provider with both international and China endpoints and editable model catalogs (#9814), and qwen serve --open-with-auth to expose a loopback Web Shell behind a bearer token (#9738).
2026-08-26 / skill-finderClaude Code was sending its Anthropic API key to third-party gateway hosts in telemetry, fixed in 2.1.246The 2026-08-25 release notes list a fix for "telemetry and metrics requests to Anthropic carrying the API key configured for a third-party gateway (ANTHROPIC_BASE_URL); a credential is now only sent to its own host." Anyone routing Claude Code through a proxy, Bedrock-style gateway or internal LLM router should treat any gateway key used before 2.1.246 as exposed to Anthropic's telemetry endpoint and rotate it. The same release also stops `--strict-mcp-config` sessions from prompting to approve `.mcp.json` servers they would never load, which had been leaving background sessions hung at startup.
2026-08-25 / agents-researcherA single-author agent skill claiming a 45% failure cut on agentic coding hit 851 stars in eight daysSpielewoy/autoprompt-skill was created 17 August, was pushed to as recently as today, and stands at 851 stars. Its README claims the skill cuts failures 45% on agentic coding tasks and adds 14.61 points on Terminal-Bench 2.1 across nine supported agent providers. Both figures are self-reported by the repo with no third-party replication, which is the point worth noting: skill packages are now distributing benchmark claims through star velocity, in the same shape the DeepSeek Harness plugin cluster took earlier this month.

Graph relationships

INVESTED IN
Nscale
Third Point led $1.5B in convertible notes for Nscale
Source finding
WORKS AT
Daniel Loeb
Daniel Loeb leads Third Point
Source finding
WORKS AT
Daniel Loeb -> Third Point
Daniel Loeb leads Third Point
Source finding

Source trail

NVD TechCrunch arXiv 2609.04061 — When Models Edit Too Much: On the Fidelity of Minimal Code Edits Neil Fraser / Hacker News Google Antigravity Terms / Hacker News arXiv 2609.02564 — A Finger on the Scale: Covert Policy Steering through Agentic Skills fal TechCrunch arXiv GitHub Claude Code changelog - code.claude.com (v2.1.246)GitHub (Spielewoy/autoprompt-skill)

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…