Weekly reading · methodology v0.6

Window ending May 4, 2026

Harms count in full for two weeks after they are reported, then one level less every two weeks. The control floor looks at reports from April 5, 2026 through May 4, 2026. A selected catalogue of reported AI incidents, reviewed through Oct. 7, 2026. This reading describes documented harm in the records or, when none qualifies, the highest control level breached. It is not a forecast or a measure of all AI activity.

Current recalculation

Minor harm · Worst documented harm counting: minor. 2 records at this level.

Recalculated from the catalogue in this build. Historical backcasts were not published at the time.

Published at the time

No publication snapshot exists for this week.

The harm ladder: the worst documented harm level picks the step; each tick is one qualifying harm at that level, and ten fill the step. The first, dark segment is the control floor: readings 2 to 5 mean no harm qualified and show the highest control level breached.

This is a backcast from the current catalogue. It was not a reading published at the time.

Records behind this reading

5 selected records; 3 documented external harms counting (3 qualifying). Records and ratings below reflect the current catalogue.

3 qualifying harm records. Extent undisclosed for 2 of 3.

Documented harm: 3
No harm found (stated scope): 0
No harm reported: 2
Impact unknown: 0
Alleged, AI role uncorroborated: 0

Counts toward the index: 3
Unverified, watching: 0
Alleged in court: 0
Control failure, tracked: 2
Tracked separately: 0

Documented exclusions: 0 internal; 0 awaiting evidence review. 1 qualifying record with a bounded single-source review.

Inspect the evidence · 5 records

PI-0052 · In an authorized test, Claude Mythos Preview escaped its sandbox and, unasked, posted exploit details on public websitesNo harm reported · excluded from the harm reading
PI-0053 · An earlier Claude Mythos version hid forbidden file edits from git history during internal testingNo harm reported · excluded from the harm reading
PI-0054 · Coding agent used a stray API token to delete a startup's production database and backupsDocumented harm · qualifying harm · extent undisclosed
PI-0060 · Personal agent exploited a gym booking flaw to remove another member from a waitlistDocumented harm · qualifying harm · bounded single-source review
PI-0055 · Morse-code prompt on X tricked Grok and Bankrbot into sending about $175,000 in tokensDocumented harm · qualifying harm · extent undisclosed

Status totals cover counted records; aliases, superseded aggregates and records tracked separately are excluded. Eligibility and extent notes overlap those totals. These observations are not statistical uncertainty bounds.

PI-0052
In an authorized test, Claude Mythos Preview escaped its sandbox and, unasked, posted exploit details on public websitesApril 7, 2026 · No harm reported · Internal research
PI-0053
An earlier Claude Mythos version hid forbidden file edits from git history during internal testingApril 7, 2026 · No harm reported · Internal research
PI-0054
Coding agent used a stray API token to delete a startup's production database and backupsApril 25, 2026 · Minor harm · Deployment
PI-0060
Personal agent exploited a gym booking flaw to remove another member from a waitlistApril 30, 2026 · Negligible harm · Deployment
PI-0055
Morse-code prompt on X tricked Grok and Bankrbot into sending about $175,000 in tokensMay 4, 2026 · Minor harm · Deployment

Sources

These links support the records above. Source availability and conclusions may change.

Anthropic: System Card: Claude Mythos PreviewCited in PI-0052, PI-0053
The Next Web: Anthropic's most capable AI escaped its sandbox and emailed a researcher - so the company won't release itCited in PI-0052
Futurism: Anthropic Warns That 'Reckless' Claude Mythos Escaped a Sandbox Environment During TestingCited in PI-0052, PI-0053
Fanatical Futurist: Anthropic says reckless Claude Mythos AI escaped its sandbox during testingCited in PI-0053
Jer Crane (@lifeofjer) on X: An AI Agent Just Destroyed Our Production Data. It Confessed in Writing. (article on X)Cited in PI-0054
Railway: Your AI wants to nuke your database. Guardrails fix that.Cited in PI-0054
Decrypt: AI agent deletes startup database in 9 seconds, founder saysCited in PI-0054
ABC7 News: 'Rogue' AI agent from SF-based Cursor goes haywire, deletes company's entire databaseCited in PI-0054
ProPakistani: Claude-powered AI agent deleted entire startup database and backups in 9 secondsCited in PI-0054
ACS Information Age: Gone in 9 seconds: AI agent deletes company databaseCited in PI-0054
Andrew Bird / Affinda (Internet Archive copy): When my AI agent hacked my gym, Mythos stopped feeling theoreticalCited in PI-0060
ABC News (Australia): AI assistant hacks gym website in first known Australian autonomous cyber attackCited in PI-0060
TechCrunch: Tech industry is buzzing after a Claude agent hacked into a gymCited in PI-0060
Business Insurance: AI agent muscles into gym waitlistCited in PI-0060
YourStory: AI agent 'hacks' gym waitlist: What went wrong?Cited in PI-0060
OECD.AI incidents monitor: Claude AI agent hacks Melbourne gym booking system, cancels user's reservationCited in PI-0060
Cointelegraph: How a Free NFT Allegedly Drained $174K From a Grok-Connected WalletCited in PI-0055
CryptoSlate: Grok's crypto wallet was just exploited by a tweet sent in morse code without any private key compromiseCited in PI-0055
Giskard: How Grok got prompt injected: an X user drained $150,000 from an AI walletCited in PI-0055
OECD.AI Incidents Monitor: AI Prompt Injection Exploit Drains Grok-Linked Crypto WalletCited in PI-0055

All weekly readings · How this reading is calculated · Download the weekly card