AI Companies Ranked by Ethics (2026): Scores and Evidence

Anthropic has the highest recorded Safe & Smart Tech score among the nine companies compared here, at +40. Apple (+20) and AMD (+10) also score positively. That corrects the earlier claim that Anthropic was the only positive company. Meta has the lowest result on this dimension, at −50.

This is a ranking of company conduct on one Mashinii dimension, not a test proving which chatbot is safest or a universal verdict on the most ethical AI company. Model developers, chipmakers and platform businesses have different roles. A purchasing decision also needs the evidence for the particular product and use case.

Which AI company is the most ethical? Define the comparison

If you mean the highest technology score in this selected group, the answer is Anthropic. If you mean privacy for a particular task, worker treatment, military exposure or climate, a different comparison is needed. An overall ethical label should not be inferred from one favourable result.

The table uses records dated 26–28 April 2026, checked on 8 October 2026. They are the latest stored ratings, not scores recalculated after every subsequent policy update. Equal technology scores share a rank.

Each company name opens its full evidence record. Higher is better on the −100 to +100 scale. A numeric zero is a recorded score, not a declaration that there is no adverse evidence or no risk. “Not scored” on a company profile means a numeric assessment is unavailable and should not be treated as zero.

What the ranking does—and does not—measure

Mashinii's methodology organises corporate evidence across eleven values. Safe & Smart Tech is broader than generative-AI model safety: a large company's privacy, security and other technology activities can affect its record.

The comparison includes three different types of business:

• Model developers: Anthropic, OpenAI and xAI.
• Chip and computing suppliers: AMD and NVIDIA.
• Broad platforms and technology groups: Apple, Alphabet, Microsoft and Meta.

Those roles overlap, but they are not interchangeable. A chip supplier's score is not a safety evaluation of every model trained on its hardware. A platform company's score is not solely a score for its newest AI assistant. This article therefore keeps the technology ranking separate from the other dimensions shown alongside it.

Anthropic leads this dimension, but its policies still need scrutiny

Anthropic's +40 technology score coexists with −20 on business conduct and −20 on climate. The table does not establish that it is positive across every value or that all its products are harmless.

Its Responsible Scaling Policy page lists revisions through version 3.4, effective July 2026, and an August risk report. Its February 2026 explanation describes a substantial policy rewrite. These are dated company disclosures: they help readers examine commitments, but publishing a policy is not independent proof that every safeguard works.

For a closer two-company comparison, read OpenAI vs Anthropic. Check the version, evaluation date and scope of any safety document before using it to support a current purchasing decision.

Apple and AMD also have positive technology scores

Apple scores +20 and AMD +10 on technology. Apple's climate and waste scores are both −20, correcting the earlier positive figures in this article. Its worker score is −10, not the previous −30.

AMD's recorded climate result is −10, business conduct +10, and weapons dimension −30. These are different questions: a positive technology score does not negate another concern, and a negative weapons score does not establish that a particular consumer chip is a weapon.

Readers comparing physical devices should also examine the product and supply chain. Our Samsung vs Apple comparison illustrates how the answer changes when workers, sourcing and device-level decisions enter the picture.

NVIDIA and OpenAI: a zero is not a clean bill of health

NVIDIA and OpenAI tie at 0 on Safe & Smart Tech. The earlier article incorrectly explained that as an absence of verified adverse conduct. A stored zero cannot support that broad factual claim.

Their other results differ. NVIDIA has 0 on business conduct, −20 on climate and −30 on weapons. OpenAI has −10, −40 and −20, respectively. Those company-level values do not measure the quality of an AI-generated answer or tell you which model will perform best in a specific workflow.

See the NVIDIA, Meta and Google scorecard for that narrower group, and Is OpenAI ethical? for the separate company assessment.

Meta, Alphabet, Microsoft and xAI: keep the order accurate

The negative technology results in this cohort are xAI −10, Microsoft −30, Alphabet −40 and Meta −50. Meta is lowest on this dimension; xAI is not the second-lowest. Alphabet's corrected −40 also changes its position relative to Microsoft.

The climate scores tell a different story: Microsoft is 0, Alphabet and Meta −10, and xAI −50. The former claims of strongly positive climate scores for Microsoft and Meta have been removed. These April records should not be interpreted as an automatic assessment of events later in the year.

An inquiry is not a finding of wrongdoing

The FTC's September 2025 AI-companion inquiry sought information from seven companies about testing, children and teens, data use and related practices. The agency describes its 6(b) authority as permitting studies without a specific law-enforcement purpose.

That announcement establishes the scope of an inquiry. It does not prove every recipient violated a law. The same care applies to lawsuits: filing an allegation, settling a claim and receiving a final judgment are different events and should be labelled accordingly.

How to choose an AI provider responsibly

Use the company scores to identify questions, then evaluate the actual product. Ask what data is retained, what testing covers your use case, how incidents are handled, and what evidence supports the provider's claims. Compare the same product category and review date.

NIST's AI Risk Management Framework provides a voluntary approach to assessing risks through design, development, use and evaluation. It is a framework for doing the work, not a certification that any company in this table is safe.

A useful decision combines the company's broader conduct with product-specific evidence. Follow the score links, inspect the cited sources and limitations, and avoid treating a single number as a substitute for that review.

Correction and review, 8 October 2026: reconciled the non-technology values and ranking order, corrected the number of positive technology scores, removed the zero-equals-no-adverse-record claim, and distinguished corporate assessments from model safety tests. Original publication date preserved.