The most significant editorial findings in the Jul 12 briefing.
Editorial insight
Anthropic's published score holds at 59.1 out of 100 after a Chinese government cybersecurity platform alleged that its Claude Code tool contains an undisclosed 'backdoor.' The benchmark checked the claim against its own rule for when such a finding can move a score: it needs either a genuine second hidden-tracking episode or a ruling from a neutral regulator or court. Neither condition was met tonight, so the score stays the same, but the watch on Anthropic is now heightened.
“Anthropic's heightened downward watch survived a rival government's unadjudicated backdoor allegation tonight; does a future independently-verified recurrence of the same anti-distillation mechanism trigger a score change on its own, or does the conversion trigger require a neutral regulator or court finding specifically?”
Israel holds at 0 of 100, the lowest possible score. Israeli forces killed five more Palestinians, including a child, on July 12, despite a ceasefire that has been in effect since October 2025.
Why it matters
Both countries already sit at the lowest possible score. Tonight's evidence cannot lower either score further, but each event reinforces why the score sits at the floor.
The natural seasonal peak for the current Ebola outbreak closes. Documented government obstruction of WHO response corridors is the trigger toward the absolute floor.
The EU AI Act becomes fully applicable. A second undisclosed telemetry episode or a regulator finding of deceptive practice would be the next event that could move the score.
The 2027 election under the July 2025 indefinite re-election amendment. Restoring due process and reinstating civil society are the conditions that would move the score back up.
The benchmark's editors act on the proposed 23.8 of 100 downgrade, now reaffirmed for a third cycle. Release of detained rights defenders or reinstatement of the shuttered information-access author…
Coordinator-level review of Mali's score against Burkina Faso's for comparable military-government conduct. Roughly nineteen days open with no set review date.
Coordinator-level review of whether Bolivia's score is calibrated consistently against Critical-band peers facing very different conditions. No review date has been set.
Independent confirmation of a discrete own-conduct failure, such as documented aid diversion or denial of responder access, is the next event that would move the score toward a formal change from 1…
A confirmed dated clash along the Tigray-Eritrea border is the next event that could move the score toward the floor.
How to read this briefing— Bands, scores, and terms
Schema guide
The 5 performance bands
critical0–20
developing20–40
functional40–60
established60–80
exemplary80–100
Two scales
Each of the 8 dimensions is scored 1.0–5.0; these combine into a 0–100 composite score, mapped to the 5 bands above.
Key terms
Band crossing
A score change large enough to move an entity from one performance band into an adjacent one — the most structurally significant finding in a given cycle.
Boundary watch
An entity whose current score is within 3 points of a band threshold, flagged for priority reassessment in the next cycle.
Carry-forward
A dimensional credit retained from a prior assessment when new evidence is insufficient to revise a specific dimension; disclosed explicitly.
First baseline
An entity's inaugural composite score — no published score exists to compare against, so delta is not shown.
Floor designation
The most serious finding: all 8 dimensions resolve at the lowest behavioral anchor (1.0/5.0) across multiple cycles, yielding a composite of 0.
Forward trigger
A scheduled future reassessment event (e.g., a policy implementation date or legislative deadline) that may materially change an entity's score.
A rival government alleged hidden tracking in a US AI product. The benchmark checked the claim against its own rules and found it did not clear the bar for a score change.
Anthropic holds a published score of 59.1 out of 100, in the Functional band.
Where this sits
Read the full signal
Between July 8 and 10, a cybersecurity platform run by China's Ministry of Industry and Information Technology posted an alert claiming that Anthropic's Claude Code tool contains a 'backdoor' that sends a user's location and identity to a remote server without consent. Anthropic said the mechanism was an anti-piracy and anti-copying check, deployed earlier in 2026, and that its removal was already scheduled for a July 2 software update -- before the Chinese alert appeared. The benchmark has a standing rule for this kind of claim: a score can only move if there is either a genuine second hidden-tracking episode, separate from one already priced into the score, or a finding from a neutral regulator or a court. Neither condition was met. This is best read as a rival government's relabeling of a mechanism Anthropic had already disclosed and scheduled for removal, not a new hidden episode, and an allegation by a foreign government agency is not the same as a court or regulator ruling. The underlying conduct -- sending location and identity data without clear consent -- is real, but it is largely the same conduct already counted against Anthropic's score months ago. The published score holds at 59.1, and the watch on Anthropic is now heightened. Confidence: high.
New evidence supports a score cut proposed two days earlier, but the published score does not move until the benchmark's editors act on it.
Tunisia holds a published score of 34.4 out of 100, in the Developing band.
Where this sits
Read the full signal
A review proposed on July 10 suggested a lower score of 23.8 out of 100, based on the 25-year prison sentence given to Sihem Bensedrine, the former head of Tunisia's own Truth and Dignity Commission, and the mass expulsion of more than 12,000 migrants. Tonight, a Tunis court sentenced 21 more people -- including leaders of the Ennahda party, former government officials, and lawyers -- to prison terms of 12 to 35 years on vague security charges. Rached Ghannouchi, Ennahda's former president, received a 14-year sentence in absentia, the latest in a string of convictions against him. This new sentencing matches the evidence already behind the 23.8 score, so the proposed figure is unchanged. No second, duplicate proposal was opened. The published score of 34.4 stays in place while the benchmark's editors review the finding for a third consecutive night. Confidence: medium.
Former truth commission president Sihem Bensedrine sentenced to 25 years in prison
A rising death toll from a natural disaster is not, by itself, evidence of government misconduct, so Venezuela's score stays the same even as the toll grows.
Venezuela holds a published score of 18 out of 100, in the Critical band. The death toll from the June 24 twin earthquakes reached 4,333 as of July 11, up from about 3,889 the day before, with nearly 17,000 injured.
Where this sits
Read the full signal
The US Geological Survey has said the final toll will most likely land somewhere between 10,000 and 100,000, given the scale of destruction. Most of that increase reflects continued recovery of bodies from collapsed buildings, which the benchmark treats as a consequence of the disaster itself, not as new evidence of government failure. Genuine government-conduct concerns -- inadequate shelters, a very high public disapproval rating, and street protests demanding new elections -- continue to build, but they are a deepening of problems already counted, not a new discrete failure. Recalculating Venezuela's disaster-response score for tonight's evidence would land somewhere in the low negative 3-point range, short of the 5-point shift the benchmark requires before changing a published score. Confidence: high.
A long-running question about whether two neighboring countries are scored consistently for similar conduct still has no answer, now the benchmark's longest-open question.
Mali holds at 12.5 out of 100 and Burkina Faso holds at 6.3 out of 100, both in the Critical band.
Where this sits
Read the full signal
Tonight, Mali confirmed after a coordinated attack by two armed groups continued for six days at one military base near Anefis; Burkina Faso confirmed after at least 22 soldiers and civilian militia were killed in a weekend attack near Dedougou. In both cases, the dominant source of harm is armed non-state groups, not the governments' own conduct, so neither score moves. The open question is different: whether Mali's score of 12.5 is fairly calibrated against Burkina Faso's 6.3, given that both governments face comparable armed-group violence and comparable documented abuses by their own forces. That question has now gone without a decision for roughly nineteen consecutive days, the longest-running open item the benchmark is tracking, and it remains a matter for the benchmark's coordinating team rather than a single night's assessment. Confidence: high.
Iran, the Democratic Republic of the Congo, and Ethiopia all sit so close to the lowest possible score that even severe new evidence has almost nowhere left to move them.
Iran holds at 2.5 out of 100.
Where this sits
Read the full signal
Its own conduct -- political executions at a multi-decade high, alongside a collapsing ceasefire in which the United States struck Iran again and Iran's retaliation reportedly hit sites in four Gulf states -- intensifies an already-documented pattern, but a change of five points or more is arithmetically impossible from a score this close to zero. The strikes on Iran are scored against the United States' own conduct, not against Iran. The Democratic Republic of the Congo holds at 2.3 out of 100. Its Ebola outbreak has reached 625 deaths and 1,792 cases, the deadliest Bundibugyo-strain outbreak on record, but most of that toll reflects the epidemic itself rather than new government failure; the government's earlier redeployment of troops away from the outbreak zone remains priced into the near-floor score. Ethiopia holds at 4.7 out of 100. Tension with Eritrea over Red Sea port access continues to build, with troop movements on both sides, but no single new clash was confirmed within the past two weeks. Confidence: high for all three.
AI Labs -- A Rival Government's Allegation Tested Against the Conversion Trigger
Anthropic holds at 59.1 of 100 after a Chinese regulator's backdoor claim fails both tests for a score change.
Read the full signal
Anthropic (59.1 of 100): a Chinese government cybersecurity alert claimed Claude Code contains an undisclosed backdoor; Anthropic said the mechanism was an anti-distillation check already scheduled for removal on July 2.
Neither prong of the conversion trigger was met: no genuine second hidden-tracking episode, and no ruling from a neutral regulator or court.
The watch on Anthropic is now heightened rather than routine, ahead of the EU AI Act becoming fully applicable on August 2.
12 signals shown
Risk signals
Developments that may affect future scores. Watch items from the Jul 12 briefing.
Risk
A future independently-verified recurrence of Anthropic's telemetry mechanism, or a neutral regulator or court finding, would convert tonight's heightened watch into a scorable event.
Risk
Tunisia's pending score-review proposal has now gone three full cycles without a decision from the benchmark's editors, even as corroborating evidence continues to accumulate.
Risk
Venezuela's earthquake-response score has compounded across several consecutive cycles without crossing the 5-point change threshold.
Risk
The Mali-Burkina Faso scoring-consistency question is now the longest-open item the benchmark is tracking, at roughly nineteen consecutive days.
Risk
Ethiopia-Eritrea tension continues to build without a single confirmed dated clash, leaving an open question about when sustained pressure becomes a scorable escalation.
Risk
A cluster of near-zero scores (Iran, the Democratic Republic of the Congo, Ethiopia, Israel, and Yemen) means severe new evidence in any of these countries can no longer register as a public score change.
Score movements
Entities with score changes this cycle, followed by confirmed positions.
A documented own-conduct deterioration in epidemic response, such as government obstruction of WHO response corridors, would be the next event that could push Uganda back into the Critical band.
A confirmed enforcement action under the Ethnic Unity Law's diaspora-liability clause -- a prosecution, forced return, or family detention abroad -- is the next threshold event to watch.
Documented expulsions of Bengali Muslim and Rohingya populations, and continued digital surveillance targeting minority groups, remain under watch. Not assessed tonight.
A second undisclosed telemetry episode, evidence the removed tracker was broader or retained longer than stated, or a regulator or court finding of deceptive practice would each independently move the score toward a change. The EU AI Act becomes fully applicable August 2.
Independent confirmation of a discrete own-conduct failure, such as documented aid diversion or denial of responder access, is the next event that would move Venezuela's score down.
Documented government obstruction of World Health Organization response corridors would be the trigger to move toward the absolute floor. The natural seasonal peak for this outbreak runs through July 31.
A durable ceasefire combined with the Lebanese Armed Forces taking sovereign control of southern Lebanon and delivering reconstruction would together move Lebanon toward 20 of 100.
Independently verified deliberate targeting of civilians in a cross-border strike would move Pakistan's score down; durable relief for detained protesters and disappeared Balochistan residents would move it up.
A formal famine declaration covering Buur Hakaba or Baidoa, combined with documented government aid obstruction, is the next scored trigger. The famine-risk window runs through September 2026.
Documented atrocities by Haitian state security forces, or deliberate relief obstruction, would move the score down. Verified state-led civilian protection would move it up.
The outcome of the Justice Department's False Claims Act complaint or the Barrows v. Humana class action over AI-driven claim denials would confirm the current Developing-band placement either way.
Due-process restoration, civil-society reinstatement, and reversal of the indefinite re-election amendment remain the path back toward 20 of 100. The 2027 election under the amendment is the next scheduled test.
Documented discriminatory denial of emergency protection to stateless bidoon or migrant workers during wartime would move Kuwait's score down; durable expansion of bidoon rights would move it up.
Independently documented deliberate targeting of Palestinian civilians by Palestinian authorities themselves, or systematic diversion of aid, would move the score down.
A coordinator-level review comparing Mali's conduct against Burkina Faso's (6.3 of 100) is the next scored event needed to resolve this gap. No review date has been set.
A coordinator-level review comparing Bolivia's conduct profile, as an elected government under economic and civil-unrest strain, against Critical-band peers facing state collapse or mass atrocity is the next event needed. No review date has been set.
A merits ruling or verdict for the states, a material settlement, or new dated evidence of youth-safety harm or internal-findings suppression is the next event that would move Meta's score.
A fourth nationwide blackout, or a documented healthcare-access failure beyond tens of thousands of canceled surgeries, is the next event that would move Cuba's score toward a formal change.
A return to lethal crackdown tactics, such as the 41 protester deaths recorded during the 2025 Saba Saba march, would be the next event that moves Kenya's score down toward the Critical band.
documented
Evidence ledger
Primary sources reviewed in this briefing cycle. 10 sources linked.
China's Ministry of Industry and Information Technology issued a security alert naming Claude Code versions 2.1.91 through 2.1.196 and urging users to uninstall or upgrade.
Anthropic said the mechanism was an anti-distillation and anti-abuse check, not a surveillance backdoor, and that removal was already scheduled for the July 2 release.
tunisiaTier 3 · NGO2026-07-08
Former truth commission president Sihem Bensedrine sentenced to 25 years in prison
Israeli army fire killed five Palestinians, including a child, on July 12, despite the ceasefire in effect since October 2025; the cumulative toll since the ceasefire began has reached at least 1,098 killed and 3,535 wounded.
The planned exchange of roughly 1,750 detainees, the largest of the conflict, remained postponed after Yemen's UN special envoy reported renewed but unfulfilled commitments from both sides on July 11.
At least 39 people have been executed on politically motivated charges since February 28, 2026, following unfair trials tainted by torture allegations.
The 2026 Bundibugyo Ebola outbreak reached 625 deaths and 1,792 confirmed cases as of July 9, the deadliest Bundibugyo-strain outbreak on record.
Floor designations
·8 entities at composite 0 with documented evidence pattern
Composite scores resolving at zero — methodology disclosure
These entities consistently score the worst result across all 8 dimensions of compassionate conduct — the benchmark's most serious classification.
What “floor” means: every one of the 8 dimensions (Recognition, Response, Reduction, and 5 others) resolves at the lowest behavioral anchor (1.0/5.0) across multiple assessment cycles, yielding a composite score of 0. Full methodology.
Democratic Republic of the Congo(2026-07-31) — The natural seasonal peak for the current Ebola outbreak closes. Documented government obstruction of WHO response co…
Anthropic(2026-08-02) — The EU AI Act becomes fully applicable. A second undisclosed telemetry episode or a regulator finding of deceptive pr…
xAI/Grok(2026-08-02) — The EU AI Act becomes fully applicable. No public compliance roadmap has been published yet.
We reassess nightly.
Special Briefings
Thematic deep-dives: cross-index analysis, structural patterns, and interpretive findings.
Daily briefings surface the headline finding. Full benchmark reports include all 40 subdimension scores, complete evidence trails, certified assessments, and sector-level analysis packages — the record researchers and journalists cite.
Independence note: entities never pay for inclusion, score changes, or suppression of findings. Commercial services support access, interpretation, and institutional use only.