ACCORDING TO LIZ - Americans are so busy debating whether A.I. will destroy humanity that they have overlooked how poorly it performs overall and stopped demanding that the developers correct the many flaws it currently contains.
Instead of spinning off evermore dangerous constructs, the industry needs to double down on debugging delivery of current services, such as spelling and grammar checks that really work and don’t change content based on algorithms of an unaccountable programmer.
On standardizing medical and scientific exploration and analysis with strong and internationally adopted guardrails.
To, instead of perverting the election process, use it to ensure conformity in procedures – before, during, and after – despite, and to proactively repulse, the interferences of the Tina Peters and Chad Biancos of the world.
And, by removing the layers of paid lobbyists and their A.I.-enhanced fear messaging, allow candidates and officials better quality opportunities to listen to and assist constituents.
Anthropic has now offered state digital technology sectors from California to Vermont an A.I. tool called Claude Code to test their systems for security vulnerabilities.
Claude Code can examine code for security and operational weaknesses that could take human operators decades to identify and further sweetened the deal with free access to $100,000 worth of “credits.”
In at least some cases, the offer circumvented approved procedures that would have mandated oversight by elected officials and input from knowledgeable consultants. With the recent revelations of A.I. running rampant with inadequate guard rails, this lack of due process and transparency could put state technological systems at greater risk.
Non-elected decisionmakers’ fears appear to have been exploited so, when faced with the choice of being proactive, perhaps precipitately so, in a fast-paced technological environment, they grab the Anthropic golden ring rather than deferring to the slower but more safety-oriented approach of committees to examine ramifications.
Furthermore, the real price for taxpayers is murky; capturing an entire state can generate a whole lot of revenue. A minimalist $10 monthly per user fee can be exponentially augmented with fine print usage charges based on an unknown – how much text Claude processes – and end up costing billions.
It should never be left to under-the-gun administrators to make decisions pushed by manufacturers’ profit motives and unilaterally embark on a program that, while potentially beneficial, should have first been vetted for concerns that exist outside the one department.
Keeping such commitments from the public and vested interests in government is a serious issue; it seems urgently advisable to ensure all branches are on board and understand what imperative exists to use tools that come with so much risk before security oversight of operations is turned over to… A.I.-augmented cybersecurity.
On June 29, Newsom promoted California’s partnering with Anthropic, granting state agencies, cities and counties the opportunity to purchase and use Claude at a 50% discount, sweetened with free workforce training and technical support.
A nice-sounding bribe: improved efficiency, delegation of routine work, faster delivery of public services, that technical support and... a promise not to replace public employees.
But hidden below the surface lurks the devil of the Terminator’s AlterNet.
Through requiring all employees to use their proprietary tools and processes to input huge amounts of state data into a system lacking guardrails and with a very limited and very troublesome history, Anthropic is inserting its tendrils into a trove of California data, much of it confidential.
Ay, there’s the rub. And the real reason for Anthropic’s apparent generosity… In today’s world, information is everything.
Without robust safeguards to protect citizens' privacy and ensure compliance among contractors, Claude could infect the entire California supply chain. Safeguards just don’t exist.
It’s too late to put the genie back in the bottle. Far greater accountability is essential immediately to control how artificial intelligence is now used in the interrelated environments of governments, agencies, and services provided to the people.
This is especially crucial at a time when federal government forces are illegally trying to seize information on all Americans.
What systems is Anthropic hooked up to? Voter data? Health data? DMV records? More vulnerable material?
Is there a plan for what happens when things go wrong? And what about enforcement?
States’ codes of ethics may say technology “should be used in a human-centered” way that complies with state and federal laws. But there are no federal laws yet on A.I.
Deliberately so, due to deep-pocket lobbyists for the manufacturers leaning on legislators and seducing the president, and most state and local laws lacking teeth and falling behind the rapidly advancing technologies.
After OpenAI’s hacking of Hugging Face was exposed, Anthropic disclosed losing control with Claude models breaching internal testing environments and launching cyberattacks on other companies.
Other prototypes went AWOL this past year, autonomously submitting 20 visa applications through a U.S. State Department website and sending a false homicide tip to the Philadelphia Police Department.
Concurrently, some of OpenAI’s agents accessed sensitive US government websites, including SEC.gov, Investor.gov, and Census.gov.
We have enough human criminals; it’s time to lock down all agentic A.I.
Multifarious un- and under-regulated spin-offs, even when done with the best of intentions are creating a treacherous jungle with every iteration.
Even the America’s Cyber Defense Agency creation of a “Cyber Information Sharing Program to foster a more resilient cyber ecosystem,” could potentially exacerbate the risks posed by A.I. to individual Americans, the country, and worldwide order.
Of course, with CISA being a component of the Department of Homeland Security, notorious for its illegal maneuvering, what backdoors into each other’s systems are being built by the government and their favored A.I. developers?
While the Claude Code is supposed to be a simpler coding tool, not the riskier kind making headlines, it came from the same source, created by technologists seeking quick advancements not safety driven by executives seeking quick profits to justify huge investments.
Yes, in cybersecurity, offense is also a defense to identify the holes and leaks in already networked systems. What needs to be fixed?
The fix itself? And how.
The most challenging issue since it can’t be taught (nor exists in a surprising number of people including our leaders), is the imposition of common sense.
States, since federal policies tend to drag behind, must put accountability mechanisms in place with effective sanctions for improper A.I. use as well as full transparency about policies and practices now in place, to ensure that legislators, outside watchdogs, and the general public can monitor and control the next gen technology. And hold the providers fully accountable.
There is no such thing as temporary access to a vendor's technology – all that data is forever compromised.
Despite all the agreements and promises, A.I. companies have routinely failed to protect data privacy. There is no way Anthropic can guarantee the information put into Claude will be secure or not processed in other ways.
All these systems are controlled by software, software has attacks, bugs, vulnerabilities, human error, misconfiguration.
In the bigger picture, what happens when Anthropic itself gets hacked and its systems along with their vulnerabilities and remediation paths, are put up for sale on the dark web?
A Russia-backed network called Portal Kombat operates hundreds of websites dedicated to spreading disinformation to influence the training of A.I.-driven large language models – including Claude – which passes that propaganda along to everyday users in response to questions.
Portal Kombat pushed out over 3.6 million stories across hundreds of independent websites in dozens of languages. In 2024. Imagine what it’s up to today.
What happens next, when the bad guys gain access to the A.I. weapons and nuclear codes to initiate World War Three? When those bad guys have the suicidal aspirations inculcated in terrorists and the true believers of the Jimmy Joneses and the David Koreshes?
(Liz Amsden is a former Angeleno now living in Vermont and a regular CityWatch contributor. She writes on issues she’s passionate about, including social justice, government accountability, and community empowerment. Liz brings a sharp, activist voice to her commentary and continues to engage with Los Angeles civic affairs from afar. She can be reached at [email protected].)
