Previdian recorded exploitation attempts against Atlassian's CVE-2026-21589 within two hours of watchTowr publishing the arbitrary file access technique. The CVSS 9.3 flaw lets an unauthenticated attacker read specific webroot files across eight Data Center products.
Perspective Coverage
6 publishers
- Builder
- Builder 27%
- Operator
- Operator 67%
- Investor
- Investor 6%
Reality
- Evidence78
- Adoption30
- Hype gap+15
- Incentives45
- Confidence70
buildOne report1 publisher One shared library spreads a 9.3 file-read flaw across all eight Atlassian Data Center products
Atlassian disclosed CVE-2026-21589, a 9.3-rated unauthenticated file-read flaw affecting all eight of its self-hosted Data Center products. A public proof of concept drew exploitation attempts within two hours of landing on 7 October.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap−5
- Incentives
- Not scored
- Confidence60
Atlassian disclosed a 9.3-rated flaw, CVE-2026-21589, that lets attackers with no login read known files in 8 self-hosted Data Center products. It wants internet-facing instances taken offline or cut off from outside access until they are upgraded or behind a blocking rule.
Perspective Coverage
15 publishers
- Builder
- Builder 24%
- Operator
- Operator 75%
- Investor
- Investor 1%
Reality
- Evidence72
- Adoption58
- Hype gap−20
- Incentives68
- Confidence68
AIR says Claude Code, Codex, GitHub Copilot and Gemini CLI all check out a pinned plugin commit without verifying they landed on it, so a trusted plugin can be swapped for attacker code during a background auto-update. Anthropic and OpenAI have patched.
Perspective Coverage
4 publishers
- Builder
- Builder 41%
- Operator
- Operator 50%
- Investor
- Investor 9%
Reality
- Evidence70
- Adoption35
- Hype gap+40
- Incentives75
- Confidence66
A fake branch name defeats the commit pin four AI coding agents rely on
Air Security's Plugin4Shell lets a plugin be swapped during a background refresh while the agent reports the audited commit. Anthropic and OpenAI have patched, Copilot has no fix, and Google is retiring Gemini CLI.
Reality
- Evidence60
- Adoption65
- Hype gap+20
- Incentives70
- Confidence58
Plugin4Shell turned a pinned commit into attacker code in four AI coding agents
Researchers at AIR found that four AI coding agents accepted a Git checkout they never verified, and because no marketplace can close the gap, the repair now travels through agent version numbers inside each enterprise.
Reality
- Evidence62
- Adoption38
- Hype gap+18
- Incentives60
- Confidence55
buildOne report1 publisher Trusting the repository author in VS Code runs the fake recruiter's task file
Microsoft's Defender Experts date the Contagious Interview campaign to December 2022. Its payload lands at the one hiring stage where a candidate is expected to clone and run a stranger's code. The same team tells employers to give staff a non-persistent VM for coding tests.
Reality
- Evidence64
- Adoption35
- Hype gap+8
- Incentives58
- Confidence62
buildOne report1 publisher Opting into native Git in SageMaker Unified Studio turns every save into a staging decision
Auto-sync put whatever sat on disk into Git history, notebook outputs and stray credentials included. File-level staging fixes that, and it also breaks every handoff a team built on the assumption that saving was enough.
Reality
- Evidence32
- Adoption
- Not scored
- Hype gap+18
- Incentives52
- Confidence34
Harness gives the coding agent its own permissions and its own audit trail
The pitch rests on a claim about your repository rather than about Harness's software, and with a free tier and click-through migration, the expensive part of testing it is the wiring you would have to move.
Reality
- Evidence22
- Adoption12
- Hype gap+46
- Incentives82
- Confidence34
A repo compromise found in week four outlives GitHub's seven-day Git event log
Wiz CIRT's poster puts 88 percent of its customers on hosted version control, where Git read and write events age out in a week and the fix, streaming them somewhere else, often sits behind a higher license tier.
Reality
- Evidence45
- Adoption30
- Hype gap+15
- Incentives80
- Confidence55
buildOne report1 publisher The agent said the Jira move failed. It had already committed the write.
A step-limit abort was reported as a task failure after the transition had gone through. The harness described its own loop, not the system of record, and a retry was one keystroke away.
Reality
- Evidence34
- Adoption12
- Hype gap+12
- Incentives38
- Confidence45
A joint police and cyber agency advisory puts fake-job crypto losses at S$15.1m. The more consequential detail is three exchange operations in four months, all running off the same tooling.
Reality
- Evidence66
- Adoption71
- Hype gap−12
- Incentives49
- Confidence63
The 2,500-org compromise was a Trivy problem. LiteLLM was the closing act.
SOCRadar's record-level data puts 95 percent of identified victims before the poisoned LiteLLM packages ever hit PyPI. Anyone who rotated only what LiteLLM touched is still exposed.
Reality
- Evidence52
- Adoption68
- Hype gap+12
- Incentives66
- Confidence55