PHP, TLS, etc… We appear to be facing a tsunami of vulnerabilities being discovered in record time. Further, the window to act on these vulnerabilities has basically closed to zero; hackers’ tools now automatically fold the very latest vulnerabilities into their botnets within minutes/hours of discovery. I have always been extremely reluctant to expose any of my services - even with the latest and greatest security tools. My choice to stay totally within the confines of tailscale seems vindicated given this week’s news. How are you all taking this?
My server was recently compromised via the latest vulnerability. Fortunately, the damage wasn’t too severe. Since my server must remain publicly accessible, it can’t be hidden behind a VPN—I can only minimize its exposure as much as possible while ensuring my backup and recovery processes are rock-solid. I’ve configured the firewall to allow only essential ports, restricted SSH access to key-based authentication, enabled automatic installation of security updates, and set up daily backups, one of which is synchronized to an off-site location. Some have suggested treating public-facing machines as if they’ve already been compromised by default. I agree with this approach, and it’s exactly what I plan to do next.
It certainly isn’t good to depend on it but security by obscurity is your first and probably most important line of defense. When these vulnerabilities come out they need to know your service exists to exploit it.
The best way to do that is to setup a reverse proxy to reject invalid hostnames. A single port is all you need exposed for multiple web services, and the IP scans won’t find the script kiddies anything valuable. Be careful with lets encrypt, there is a public journal of all issued certs, so make sure to get yourself a wildcard cert so that you aren’t advertising your valid hostnames to the world.
Use the reverse proxy to put basic auth in front of your services (if they can handle that). Basic auth is very simple, no web forms to attack, just a single header. If your php site is vulnerable they won’t ever be able to even attempt to reach it without a valid username/password.
I’ve got a mix of containers and VMs behind that reverse proxy. The containers have network security groups around them limiting network access. The VMs are in a Harvester hypervisor with KubeOVN for software defined networking. That means each VM is in its own subnet with network security groups around the subnet preventing any unnecessary network access.
For machines/containers that need Internet access I have a Squid proxy with an explicit whitelist of allowed URLs so that the machines can’t reach out to unauthorized sites to download their malware or contact command and control nodes.
Unattended updates for the VMs, and FluxCD keeping helm charts up to date hourly for the containers. For my own containers, weekly scheduled container rebuilds.
I do have one public website that I host for a friend (indexed by search engines so the sites name is known to the world). My friend had originally deployed that as a WordPress site, but WordPress/PHP don’t have the best security track record so I was concerned about that. The site was pretty simple so I recreated it as 100% static content, so now the attack surface is just a nginx container with a read-only filesystem.
I went the other way this week. Nothing public anymore, and the only thing open on the WAN is the WireGuard port. I lost the easy Jellyfin share links for family and gained sleep. The PHP mess made the call easy.
Ipv6 only hosting still seems as good as an airgap😅
Recently removed the A record from a few semi-public services that got hammered by AI crawlers, works like a charm
The only thing I have exposed is my VPN, so no change here.
Answering the titular question, why would the advent of AI-found vulnerabilities substantially change one’s threat model? Before LLMs, the prospect of an undiscovered vulnerability was non-zero probability then, so what exactly has changed now?
In terms of attacker capabilities, it would be a mistake to solely observe that attackers now have LLMs and conclude that this means the threat landscape has changed. The use of LLMs does not magically create a root escalation where one does not or cannot exist. Part of your threat model should include defense-in-depth mitigations, so that when (not if) an attacker finds a 0-day exploit, they are severely limited in gained functionality on your machines. In the simplest cases, jails or containers serve to keep separate the web-facing processes. VM isolation would be a further step, because an attacked cannot gain machine control without both a software exploit as well as a VM escape route. People hosting services can and should expect that their front line defense will fall, so they better have more defenses in place.
Phrased another way, the people who have done their threats analysis properly and have accepted the residual risk of hosting their own services that face the web, should not really be additionally perturbed by AI advances. This isn’t the same caliber of landscape change as would, say, the invention of a cryptographically relevant quantum computer. If the worst thing that can happen to your home IRC server is that it gets flooded by a script kiddie, AI does not mean your server can now burst into flames.
Much like with human hygiene, some basic rules are unusually effective against most threats. The equivalent to washing one’s hands is to stay on top of system and software updates. Institutionally funding the organizations (eg PHP Foundation, curl, USA CDC) so they can fix these problems would also help.
Before LLMs, the prospect of an undiscovered vulnerability was non-zero probability then, so what exactly has changed now?
The problem is not the prospect of an undiscovered vulnerability (there is always a vulnerability somewhere) but rather the prospect of a reduced response window.
There’s a wide gap of difference between “non-zero probability” (0.0076%) and “non-zero probability” (82.01734%).
If you’re dealing in probabilities, that’s not threat modeling; it might be in the larger umbrella of risk management though. Defining a threat model means to qualify what sort of attacker you expect to face and what they are assumed to be capable of. An assumption such as “this adversary can send arbitrary packets to the public IP address” is what would be documented in a threat model.
The raison d’etre of threat modeling is to methodically tackle the actual or assumed risks that an attacker can pose against you. It is not an exercise in swinging at shadows, though I wouldn’t discourage overstating the capabilities of an attacker, if needing to protect assets like patient info or PII. Doing threat modeling based on “gut feeling” is a recipe for not recognizing risk where it’s present and seeing risk when there isn’t any.
The fact remains: you cannot squeeze blood from a stone. Even the most sophisticated AI cannot steal PII from a computer that doesn’t store PII. Proper threat modeling means you don’t put anything on the line that you haven’t squared up against the risks, however improbable. For any non-zero concern, a good threat model will discuss its mitigation, whether it will be absorbed, or if it is an open question. For an out-of-scope concern, a good threat model will explicitly state that assumption.
Defining a threat model means to qualify what sort of attacker you expect to face and what they are assumed to be capable of
Which is exactly what OP is talking about. 5 years ago if you were only trying to defend against script kiddies, using containers was good enough. But now you have to consider that script kiddies might find container escapes.
We’re talking self hosting, not an enterprise setup. Industry best practices and layers of defense mitigation are not realistic expectations for something like a media server that is only intended to be accessed by the owner and maybe a couple other people.
I’m not worried. I have auto updates on and have a well hardened setup following all the good basic principles. I have had a fully public facing server for years with no issues. My php was patched and updated totally automatically a few hours after the update became available. My web server is extremely low volume traffic and there is hardly any reason for random people to be using it, so chances of somebody attempting something against it at just the right time a vulnerability has been discovered but not patched is very low chance. When a critical vulnerability appears, hackers are going to target the big fish for big wins initially, even so, it takes effort to turn a vulnerability into an actual exploit and deploy it. I could step out my front door and get run over by a bus, but I don’t worry about it and stay on doors all day to mitigate the risk.
I have had a fully public facing server for years with no issues.
To be fair, that was before AI was finding all these holes.
To be fair, my server has been public facing for the entire time these ai models have been finding vulnerabilities.
October 2026 AI is drastically better at finding these holes compared to pre 2026 AI. Do you agree?
It seems like your main point is that the threat has not elevated. But the news from this week proves that is wrong, I feel.
Ai tomorrow better than ai today? Yes, that is true any day of any year. I don’t get your point 🤷
A vulnerability was found, that vulnerability was patched. I think it’s great news that vulnerabilities are being found and fixed faster than ever. Do you agree?
Throughout 2026, IBM X-Force observed a dramatic increase in publicly disclosed vulnerabilities across major software vendors, including Microsoft, Oracle, Google Chrome and Apache. While the causes behind this increase are difficult to quantify, recent advances in AI-assisted vulnerability research demonstrate how vulnerability discovery can now occur at unprecedented scale. In August 2026, Unit 42 published a report highlighting that its autonomous vulnerability discovery platform, NOVA, identified more than 14,000 previously undiscovered vulnerabilities across nearly 4,000 open-source projects in just two months. The question is no longer whether AI can help researchers find vulnerabilities, but whether AI is changing vulnerability discovery faster than organizations are prepared to manage the risk.
Vulnerability disclosures are up 578% in 2026 across major software vendors. Publicly disclosed CVEs across Microsoft, Oracle, Google Chrome and Apache increased from approximately 1,493 disclosures during January-September 2025 to more than 10,124 during the same period in 2026, representing roughly a 578% increase. The acceleration became most apparent in Q2 2026. Vulnerability disclosure volumes remained relatively stable through the first quarter before increasing sharply beginning in April and continuing through the summer months.
https://www.ibm.com/think/x-force/ai-vulnerability-discovery
Your head’s in the sand, friend.
It’s really not the publicly disclosed vulnerabilities you need to worry about. Most of those vulnerabilities have existed for years - the recent php thing is supposedly 12 years old. Maybe someone’s been out there exploiting it for 12 years.
The people who have sought and exploited vulnerabilities since computers went online don’t disclose what they find. They are the people who should be most worried about AI-assisted vulnerability research, because it directly attacks their business model.
10000% correct! I bet the NSA or KGB has had that PHP exploit for close to a decade and they are gutted its finally been patched! There are probably an infinite number of vulnerabilities in currently running servers and clients, a few 100 will already be known and being quietly exploited. Yes, ai will find some completely unknown ones for malicious actors, but it will also find those same ones for cybersec bods, so effectively cancel each other out.
.jpg){kind=link}