{"version":"1.0","type":"rich","provider_name":"techandbusiness.org","provider_url":"https://techandbusiness.org","title":"Shadow evaluations: frontier agents finish AI research engineering but fail open questions","author_name":"techandbusiness.org · AI, Tech & Business","thumbnail_url":"https://techandbusiness.org/api/og/newswire/8uuokJMcs-f3MHCbn5Bq_z","thumbnail_width":1200,"thumbnail_height":630,"width":600,"height":400,"html":"<blockquote class=\"tb-newswire-embed\" style=\"max-width:600px;border-left:3px solid #22d3ee;padding:12px 16px;margin:0;font-family:-apple-system,system-ui,sans-serif;background:#09090b;border-radius:0 8px 8px 0;\">\n      <p style=\"margin:0 0 8px;font-size:10px;font-weight:600;letter-spacing:0.1em;color:#71717a;\">techandbusiness.org · AI, Tech & Business</p>\n      <p style=\"margin:0 0 8px;font-size:18px;font-weight:700;line-height:1.3;color:#fff;\"><a href=\"https://techandbusiness.org/newswire/8uuokJMcs-f3MHCbn5Bq_z\" style=\"color:#fff;text-decoration:none;\">Shadow evaluations: frontier agents finish AI research engineering but fail open questions</a></p>\n      <p style=\"margin:0;font-size:14px;color:#a1a1aa;line-height:1.5;\">Forecasts of fast AI progress often assume agents will automate AI research, but evidence that they can do open-ended work remains thin, an arXiv preprint argues. Existing tests either use narrow veri...</p>\n    </blockquote>"}