buildConfirmed3 publishers Zenity Labs says one prompt to a public Bedrock AgentCore agent returned credentials for a role scoped to every agent in the account and region. AWS had removed the wider default permissions by September 29, 2026, but roles a team copied or widened sit in its own account.
Perspective Coverage
4 publishers
- Builder
- Builder 41%
- Operator
- Operator 41%
- Investor
- Investor 18%
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+30
- Incentives60
- Confidence58
buildOne report1 publisher Confirmed Claude-found open source bugs outnumber upstream patches 11 to 1
Anthropic's Claude flagged more than 29,000 candidate open source vulnerabilities in six months, and upstream projects have patched 516 of them. Maintainers praise the reports, but each fix still ships only as fast as the project that tests and releases it.
Reality
- Evidence55
- Adoption45
- Hype gap+15
- Incentives60
- Confidence55
buildOne report1 publisher Telegram Desktop's missing semicolon escape turned a clicked link into session theft
Telegram Desktop 7.2.9 patches CVE-2026-107181, a CVSS 8.1 flaw where one clicked link in a group chat could steal a victim's logged-in account session. Every build up to 7.2.8 is vulnerable, so the fix helps a machine only once the patched client is installed.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence40
buildOne report1 publisher GROWI 7.5.5 closes an anonymous file read on wikis that store uploads locally
GROWI, Inc. shipped GROWI 7.5.5 to close CVE-2026-100727, a flaw letting visitors with no account read files from non-public pages on local-upload wikis. Because the read needs no login, any reachable Local-mode instance on an older release is a patch-now item.
Reality
- Evidence55
- Adoption30
- Hype gap+5
- Incentives
- Not scored
- Confidence50
Anthropic expects more than 90% of the unreviewed vulnerability reports its new OSS Scanner sends to open source maintainers to be real. Opt-in projects get findings faster and take on the work of catching the errors, such as wrong severity ratings.
Perspective Coverage
6 publishers
- Builder
- Builder 52%
- Operator
- Operator 37%
- Investor
- Investor 11%
Reality
- Evidence60
- Adoption35
- Hype gap+15
- Incentives70
- Confidence62
Researchers at Pwn2Own Ireland 2026 collected $1,262,000 for exploiting 98 zero-day flaws across seven product categories in three days. Every exploit had to work on current firmware, so owners of those products have to wait for vendor fixes.
Reality
- Evidence70
- Adoption
- Not scored
- Hype gap+10
- Incentives45
- Confidence72
Exploited Zammad bug leaks helpdesk session cookies through Ruby error output
CISA added Zammad's unauthenticated WebSocket session-disclosure flaw, CVE-2026-102489, to its exploited-vulnerabilities catalog on October 2. The fix ships in Zammad 7.2.0, leaving teams on the unsupported 6.5-and-earlier releases to make a major-version jump.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+5
- Incentives70
- Confidence55
Commercial AI models refused every query from Hugging Face's breach responders, Veracode's Chris Wysopal wrote, forcing them onto a self-hosted Chinese model. Security leaders now have to settle which AI model their responders can use before an intrusion starts.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+25
- Incentives40
- Confidence50
Pwn2Own Ireland teams logged 32 zero-days and over $368,000 in prizes on October 6, breaking LiteLLM, OpenAI Codex and Oracle's AI database. Vendors now have 90 days to ship fixes before the Zero Day Initiative publishes the details.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+10
- Incentives40
- Confidence62
Pwn2Own Ireland's first day puts LiteLLM and OpenAI Codex among 32 exploited zero-days
Pwn2Own Ireland 2026 researchers exploited 32 zero-days on day one for $388,500, including LiteLLM flaws and an OpenAI Codex argument injection. Vendors now have 90 days to ship fixes before ZDI publishes details, and the report describes contest exploits only.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+10
- Incentives35
- Confidence55
DIVD, the Dutch volunteer disclosure group, was breached through two chained zero-days in its own Zammad helpdesk that took an attacker to root in seconds. Zammad claims over 2,000 customers, and DIVD wants every older install upgraded to version 7 or taken offline.
Perspective Coverage
9 publishers
- Builder
- Builder 31%
- Operator
- Operator 58%
- Investor
- Investor 11%
Reality
- Evidence70
- Adoption40
- Hype gap+25
- Incentives45
- Confidence65
buildOne report1 publisher Anthropic's red team finds open-source bugs faster than maintainers can ship fixes
Anthropic's Frontier Red Team has disclosed 6,157 open-source vulnerabilities and knows of upstream patches for 516. The gap sits with maintainers, whose capacity Anthropic's own policy treats as the ceiling on how fast it files reports.
Reality
- Evidence55
- Adoption20
- Hype gap+15
- Incentives55
- Confidence50
buildOne report1 publisher Four chained bugs let unauthenticated attackers seize OpenBao and Vault servers with Raft snapshot policies
OpenBao fixed a four-bug chain in 2.6.3 and 2.7.0 that lets unauthenticated attackers seize servers with a Raft snapshot policy set. A dev.to post says HashiCorp Vault has no fix yet, so Vault operators have to close network paths to the API themselves.
Reality
- Evidence22
- Adoption
- Not scored
- Hype gap+45
- Incentives
- Not scored
- Confidence25
Missing token signature check gave a 16-year-old admin SQL on a Microsoft analytics service
Faav, a 16-year-old researcher, reached admin SQL over 17.3 trillion rows on Microsoft's Titan service because it never checked login-token signatures. Microsoft locked the endpoint four days after his report and paid a $5,000 bounty.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+35
- Incentives55
- Confidence55
buildConfirmed2 publishers Copilot Autofix swapped an env-var-and-jq pattern for inline interpolation in a public GitHub Actions file, and Wiz's autonomous agent exploited it inside a week.
Reality
- Evidence72
- Adoption
- Not scored
- Hype gap+35
- Incentives65
- Confidence62
Adversa AI says its Cryptographic Context Injection recovers hostile prompts inside the code sandbox, where filters do not look. xAI has not replied; Google scopes jailbreaks out entirely.
Perspective Coverage
6 publishers
- Builder
- Builder 34%
- Operator
- Operator 55%
- Investor
- Investor 11%
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+30
- Incentives55
- Confidence55
CVE-2026-75501 puts an unauthenticated UPnP control endpoint on the WAN side of a premium ISP gateway. With no patch and no vendor reply, the carriers own the mitigation.
Reality
- Evidence62
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence58
Olivier Laflamme walked an unpaired Bluetooth write up to root on the G1 EDU's Locomotion PC by way of a Unitree cloud API that decrypted key material for any logged-in account. Unitree fixed that check in July, but no patched firmware has been named.
Perspective Coverage
4 publishers
- Builder
- Builder 38%
- Operator
- Operator 50%
- Investor
- Investor 12%
Reality
- Evidence72
- Adoption38
- Hype gap+8
- Incentives45
- Confidence62
CERT Polska dated successful attacks to at least September 2 and published its warning on September 5, so operators who deferred the RouterOS update have three days of configuration changes to read as well as a patch to install.
Perspective Coverage
8 publishers
- Builder
- Builder 19%
- Operator
- Operator 73%
- Investor
- Investor 8%
Reality
- Evidence78
- Adoption45
- Hype gap+18
- Incentives30
- Confidence72
MikroTik published RouterOS fixes in four branches with no CVE and no technical detail. The same upgrade runs a compromise check and writes a critical log entry marking the device Flagged.
Perspective Coverage
3 publishers
- Builder
- Builder 22%
- Operator
- Operator 71%
- Investor
- Investor 7%
Reality
- Evidence76
- Adoption
- Not scored
- Hype gap−40
- Incentives45
- Confidence72