Agentic market $10.8B and climbing  ·  editor@gaasnews.com
Sections
HomeWhat is GaaS?PlatformsPricingGlossaryOpinionAboutContact
HomeEvaluation & SafetyAI Safety Index: Best Grade Is a C+
Evaluation & Safety

Nine AI companies got a safety report card. The best grade was a C+

The Future of Life Institute's summer index puts Anthropic on top, fails a third of the industry outright, and flunks everyone on existential safety.

AJ
Andrew Jamerson
Founding Editor
Jul 8, 2026 · 3 min read
Woman in hoodie with technology brain conceptPhoto: Woman in hoodie with technology brain concept. // GaaS News
TL;DR
  • The Future of Life Institute published its Summer 2026 AI Safety Index on July 7, grading nine frontier AI companies with a seven-member expert panel.
  • Anthropic leads at C+ (2.66), followed by OpenAI at C (2.28) and Google DeepMind at C (2.01). Meta gets a D+, and xAI, DeepSeek and Mistral all receive an F.
  • On existential safety, no company scored above a C-. Six of nine got an F in that domain.
  • Reviewers also note companies have weakened or dropped earlier pledges to pause at capability red lines.

The companies whose models power the agent economy just received their semiannual report card, and the best grade in the class is a C+. The Future of Life Institute's Summer 2026 AI Safety Index, assembled by a seven-member expert panel that includes UC Berkeley's Stuart Russell, the University of Montreal's David Krueger and Yi Zeng, grades nine firms across domains from risk assessment to governance.

The grades

Anthropic tops the table at C+ with a 2.66 score, with OpenAI at C (2.28) and Google DeepMind at C (2.01) behind it. Meta lands a D+, Z.ai and Alibaba Cloud each take a D-, and xAI, DeepSeek and Mistral all fail outright. The hardest domain was the bluntest: on existential safety, planning for systems that could escape meaningful human control, nobody cleared a C-, and six of the nine companies received an F.

The retreat from red lines

The panel's sharpest finding is directional, not numerical. Time reports that companies across the index have weakened or abandoned earlier commitments to halt or pause when capabilities crossed danger thresholds. Russell's verdict: "Companies have backed away from earlier commitments to release new systems only with safety measures appropriate for their capability levels; now, they're planning to release them even if it's demonstrably unsafe to do so." Krueger called the industry's lack of credible safety planning "scandalous."

Why this lands on the GaaS desk

Every outcome-priced agent this publication covers runs on one of these graded models. When an enterprise buys resolutions, filings or research from an agent vendor, it inherits the safety posture of whichever lab sits underneath, usually without seeing it. That makes the index something more practical than advocacy: it is a procurement document. Buyers negotiating agent contracts now have a citable, expert-graded baseline for asking vendors which model they run on and what its safety grade is. Pair it with the BeSafe agent benchmark, where no agent cleared 40% on safe task completion, and the pattern is consistent: capability is racing ahead of control, and the market is pricing capability.

Sources: Future of Life Institute, Time, Axios.

Last fact-checked: Jul 8, 2026 by Andrew Jamerson
AJ

Andrew Jamerson

Founding Editor, GaaS News

Andrew Jamerson is the founding editor of GaaS News, covering the economics of the agent era. He started the publication to cover Agentic AI as a Service as a dedicated beat and edits every article on the site.

Be on the list when the beat breaks

One email when a platform ships, a round closes, or the ground shifts under the software stack.