Memory regression after 2.1.227/desktop 1.28929.0: 3-6x footprint growth causes OOM SIGKILLs misreported as "Blocked by endpoint security"

Summary

After updating to Claude Desktop 1.28929.0 / Claude Code 2.1.227 (both installed Aug 11-12), per-process memory footprint grew 3-6x. On a 24 GB Mac this drives free RAM low enough that macOS SIGKILLs Claude Code sessions in batches. The desktop app then shows a "Blocked by endpoint security" dialog naming Team ID Q6L2SF6YDW, which misattributes the kill - no endpoint security is active on this machine.

Environment

• macOS 26.5.2 (arm64), 24 GB RAM
• Claude Desktop 1.28929.0 (bundle modified Aug 11 19:06)
• Claude Code 2.1.227 (installed Aug 12 10:33, only version on disk)
• Endpoint security: none active (FortiClient present but unregistered, VPN extension only; binary passes codesign + Gatekeeper)

Evidence: before vs after (same [process-memory] telemetry in ~/Library/Logs/Claude/main*.log)

| Metric | Before update (main1.log, through Aug 12 07:53) | After (main.log, Aug 12 18:00+) |
|---|---|---|
| Claude tree RSS total | 2,188-2,879 MB | 6,800-9,200 MB |
| Child process count | 140-146 | 60-86 |
| Largest renderer | ~300 MB | 1,547-1,634 MB |
| Largest CC session | ~246 MB | 1,532-1,755 MB |
| electron_main | ~238 MB | 679-878 MB |
| SIGKILL events | 0 in 4,741 samples | 65 |

Key point: before the update the machine ran with more child processes (140+) and as little as 56 MB raw free RAM with zero kills. After the update, half the process count uses 3x the memory.

First SIGKILL ever logged: 2026-08-12 18:01:10 - ~23 hours after the desktop bundle updated. 65 events across 19 batches since.

Kill signature (why we believe this is OOM, not EPP)

Kills arrive 3-6 within the same second, always following a sample where sys_free_raw collapses:

12:41:05 tree_rss_sum=9204MB children(86) sys_free_raw=514MB
12:43:00 <- 6x "Claude Code process terminated by signal SIGKILL"
12:43:05 tree_rss_sum=4552MB children(39) sys_free_raw=2657MB

Tree RSS drops ~4.6 GB and child count halves immediately after each batch. Freeing system memory (quitting other apps) stops the kills entirely.

Renderer leak

The main window renderer grows ~400 MB/hour: 298 MB at app launch (10:39) -> 1,634 MB two hours later. Restarting the app resets it; growth resumes.

Two issues

1. Memory regression / leak: 3-6x per-process footprint growth after the Aug 11-12 update, plus continuous renderer growth (~400 MB/h).
2. Misleading error dialog: SIGKILL from the kernel OOM path is reported as "Blocked by endpoint security - ask your IT team to allowlist Team ID Q6L2SF6YDW". On machines with no EPP this sends users down entirely the wrong debugging path. Suggest the dialog check memory pressure telemetry (already collected by [process-memory]) before blaming endpoint security.

Repro

1. 24 GB Mac, Claude Desktop 1.28929.0 + Claude Code 2.1.227
2. Open 4+ concurrent Claude Code sessions with a typical MCP roster; leave the app running a few hours
3. Watch [process-memory] lines in ~/Library/Logs/Claude/main.log - when sys_free_raw drops below ~500 MB, sessions are SIGKILLed in batches and the endpoint-security dialog appears