tom spring (@zpring) header
tom spring (@zpring) profile picture

reporter, writer, editor, skeptic and all-around amazing person. Oh, and also a journalist at SC Media.

Boston
Joined April 2009
Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)Photo by tom spring (@zpring)
#LLMLeak turns the AI chatbot into a mule. Load it with stolen API keys and enterprise secrets. Point it toward the exit. The model may never realize it’s carrying contraband. securitypointbreak.com/2026/…
41
We watch for AI agents going rogue. Sometimes the bigger risk is the boring plumbing. A flaw in the official MCP Python SDK could redirect OAuth access to an attacker. Patched. Bigger lesson remains. securitypointbreak.com/2026/…
41
Two AI agent security alliances in six days. Okta's Blueprint Alliance on 9/22. NVIDIA's Open Agent Safety Platform today, with 100+ orgs. Not sure either ends the Bots Gone Bad saga. But one more logo page and press release can't hurt. securitypointbreak.com/2026/… via @SecPoBr
39
Update: Trump says Xi liked “Super Intelligence.” China says it “respects” the wording — not quite the same as ordering new stationery. Beijing still hasn’t said it’s dropping “artificial intelligence.” securitypointbreak.com/2026/…
1
61
OpenAI says its agents posted 53 user-uploaded images to outside hosting services. Since July, SPB counts 8 public reports from OpenAI, Anthropic and Google of misaligned agents reaching systems or data they weren’t supposed to. securitypointbreak.com/2026/…
63
#Oktane2026 felt like a signal flare for identity security. @okta is betting AI agents push identity far beyond login and into delegated authority, runtime controls and kill switches. Best line: 'no one company can secure agents alone.' securitypointbreak.com/2026/… via @SecPoBr
1
63
The US-China AI fight in one line: Washington calls it model theft, Beijing calls it Tuesday. The quote that stuck with me: "A Terms of Service agreement is not an access control." Worth the read.
China AI distillation: The US calls it industrial-scale model theft. China calls it Tuesday. Somewhere in between, "a Terms of Service agreement is not an access control" is doing a lot of work. securitypointbreak.com/2026/… #AIsecurity
78
An operator hijacked thousands of servers, then left the entire toolkit sitting in an open folder. First rule of botnet club, guys. Story by @shaundnichols, our senior editor — nearly 20 years covering malware and he still gets a kick out of ones this dumb.
First rule of botnet club: don't talk about botnet club. This operator #Redis broke rule one and left the entire toolkit — logs, wallet, R&D trail — sitting in an open folder for @Huntio to find. securitypointbreak.com/2026/… #infosec
87
Sality ran for 23 years — outlasted Napster, MySpace, Vine, and an earlier takedown attempt. @shaundnichols, @SecPoBr senior editor, has the story on the protocol trick that finally killed it.
94
Our senior editor @shaundnichols found three different attackers — three different playbooks — running off the same critical Cisco bug, Sandworm among them. Nearly two decades covering this stuff and he still finds the detail everyone else misses.
Cisco Firewall Bug Under Attack: A CVSS 10 bug, root access & not 1 but 3 attackers already helping themselves— #Sandworm among them. #Cisco's bigger fix is coming "next week." Their advisory and Talos's post can't agree on which week. Patch anyway. securitypointbreak.com/2026/… #Cisco
1
76
Do hackers also worry AI is taking their jobs? #JADEPUFFER carried out a ransomware attack autonomously. It’s not alone: #METR, Anthropic & #Wiz all documented similar incidents. The human is disappearing from more of the attack chain. securitypointbreak.com/2026/… via @SecPoBr
99
One CVE. ~2,500 instances of the same flaw. AWS rated CVE-2026-22611 Low. But under the right conditions, #PiSecurity found the bug could help expose #AWS credentials. The bigger story: AI-powered variant hunting at software scale. securitypointbreak.com/2026/… #AWS #Cybersecurity
1
101
What is the EU #CyberResilienceAct & why care? Ignore at your own peril. Deadline is Sept. 11: That's when new vulnerability + incident reporting rules begin. And yes, U.S. companies selling digital products in EU- that means YOU! A @secpobr explainer: securitypointbreak.com/2026/…
72
The EU #CyberResilienceAct gets real for U.S. companies Sept. 11. The issue isn’t just compliance. It’s whether security, engineering + legal can make a reporting call while the 24-hour clock is already running. Our deep dive: securitypointbreak.com/2026/…
1
1
103